xref: /linux/arch/powerpc/kernel/prom_init.c (revision da1d9caf95def6f0320819cf941c9fd1069ba9e1)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Procedures for interfacing to Open Firmware.
4  *
5  * Paul Mackerras	August 1996.
6  * Copyright (C) 1996-2005 Paul Mackerras.
7  *
8  *  Adapted for 64bit PowerPC by Dave Engebretsen and Peter Bergner.
9  *    {engebret|bergner}@us.ibm.com
10  */
11 
12 #undef DEBUG_PROM
13 
14 /* we cannot use FORTIFY as it brings in new symbols */
15 #define __NO_FORTIFY
16 
17 #include <linux/stdarg.h>
18 #include <linux/kernel.h>
19 #include <linux/string.h>
20 #include <linux/init.h>
21 #include <linux/threads.h>
22 #include <linux/spinlock.h>
23 #include <linux/types.h>
24 #include <linux/pci.h>
25 #include <linux/proc_fs.h>
26 #include <linux/delay.h>
27 #include <linux/initrd.h>
28 #include <linux/bitops.h>
29 #include <linux/pgtable.h>
30 #include <linux/printk.h>
31 #include <linux/of.h>
32 #include <linux/of_fdt.h>
33 #include <asm/prom.h>
34 #include <asm/rtas.h>
35 #include <asm/page.h>
36 #include <asm/processor.h>
37 #include <asm/interrupt.h>
38 #include <asm/irq.h>
39 #include <asm/io.h>
40 #include <asm/smp.h>
41 #include <asm/mmu.h>
42 #include <asm/iommu.h>
43 #include <asm/btext.h>
44 #include <asm/sections.h>
45 #include <asm/machdep.h>
46 #include <asm/asm-prototypes.h>
47 #include <asm/ultravisor-api.h>
48 
49 #include <linux/linux_logo.h>
50 
51 /* All of prom_init bss lives here */
52 #define __prombss __section(".bss.prominit")
53 
54 /*
55  * Eventually bump that one up
56  */
57 #define DEVTREE_CHUNK_SIZE	0x100000
58 
59 /*
60  * This is the size of the local memory reserve map that gets copied
61  * into the boot params passed to the kernel. That size is totally
62  * flexible as the kernel just reads the list until it encounters an
63  * entry with size 0, so it can be changed without breaking binary
64  * compatibility
65  */
66 #define MEM_RESERVE_MAP_SIZE	8
67 
68 /*
69  * prom_init() is called very early on, before the kernel text
70  * and data have been mapped to KERNELBASE.  At this point the code
71  * is running at whatever address it has been loaded at.
72  * On ppc32 we compile with -mrelocatable, which means that references
73  * to extern and static variables get relocated automatically.
74  * ppc64 objects are always relocatable, we just need to relocate the
75  * TOC.
76  *
77  * Because OF may have mapped I/O devices into the area starting at
78  * KERNELBASE, particularly on CHRP machines, we can't safely call
79  * OF once the kernel has been mapped to KERNELBASE.  Therefore all
80  * OF calls must be done within prom_init().
81  *
82  * ADDR is used in calls to call_prom.  The 4th and following
83  * arguments to call_prom should be 32-bit values.
84  * On ppc64, 64 bit values are truncated to 32 bits (and
85  * fortunately don't get interpreted as two arguments).
86  */
87 #define ADDR(x)		(u32)(unsigned long)(x)
88 
89 #ifdef CONFIG_PPC64
90 #define OF_WORKAROUNDS	0
91 #else
92 #define OF_WORKAROUNDS	of_workarounds
93 static int of_workarounds __prombss;
94 #endif
95 
96 #define OF_WA_CLAIM	1	/* do phys/virt claim separately, then map */
97 #define OF_WA_LONGTRAIL	2	/* work around longtrail bugs */
98 
99 #define PROM_BUG() do {						\
100         prom_printf("kernel BUG at %s line 0x%x!\n",		\
101 		    __FILE__, __LINE__);			\
102 	__builtin_trap();					\
103 } while (0)
104 
105 #ifdef DEBUG_PROM
106 #define prom_debug(x...)	prom_printf(x)
107 #else
108 #define prom_debug(x...)	do { } while (0)
109 #endif
110 
111 
112 typedef u32 prom_arg_t;
113 
114 struct prom_args {
115         __be32 service;
116         __be32 nargs;
117         __be32 nret;
118         __be32 args[10];
119 };
120 
121 struct prom_t {
122 	ihandle root;
123 	phandle chosen;
124 	int cpu;
125 	ihandle stdout;
126 	ihandle mmumap;
127 	ihandle memory;
128 };
129 
130 struct mem_map_entry {
131 	__be64	base;
132 	__be64	size;
133 };
134 
135 typedef __be32 cell_t;
136 
137 extern void __start(unsigned long r3, unsigned long r4, unsigned long r5,
138 		    unsigned long r6, unsigned long r7, unsigned long r8,
139 		    unsigned long r9);
140 
141 #ifdef CONFIG_PPC64
142 extern int enter_prom(struct prom_args *args, unsigned long entry);
143 #else
144 static inline int enter_prom(struct prom_args *args, unsigned long entry)
145 {
146 	return ((int (*)(struct prom_args *))entry)(args);
147 }
148 #endif
149 
150 extern void copy_and_flush(unsigned long dest, unsigned long src,
151 			   unsigned long size, unsigned long offset);
152 
153 /* prom structure */
154 static struct prom_t __prombss prom;
155 
156 static unsigned long __prombss prom_entry;
157 
158 static char __prombss of_stdout_device[256];
159 static char __prombss prom_scratch[256];
160 
161 static unsigned long __prombss dt_header_start;
162 static unsigned long __prombss dt_struct_start, dt_struct_end;
163 static unsigned long __prombss dt_string_start, dt_string_end;
164 
165 static unsigned long __prombss prom_initrd_start, prom_initrd_end;
166 
167 #ifdef CONFIG_PPC64
168 static int __prombss prom_iommu_force_on;
169 static int __prombss prom_iommu_off;
170 static unsigned long __prombss prom_tce_alloc_start;
171 static unsigned long __prombss prom_tce_alloc_end;
172 #endif
173 
174 #ifdef CONFIG_PPC_PSERIES
175 static bool __prombss prom_radix_disable;
176 static bool __prombss prom_radix_gtse_disable;
177 static bool __prombss prom_xive_disable;
178 #endif
179 
180 #ifdef CONFIG_PPC_SVM
181 static bool __prombss prom_svm_enable;
182 #endif
183 
184 struct platform_support {
185 	bool hash_mmu;
186 	bool radix_mmu;
187 	bool radix_gtse;
188 	bool xive;
189 };
190 
191 /* Platforms codes are now obsolete in the kernel. Now only used within this
192  * file and ultimately gone too. Feel free to change them if you need, they
193  * are not shared with anything outside of this file anymore
194  */
195 #define PLATFORM_PSERIES	0x0100
196 #define PLATFORM_PSERIES_LPAR	0x0101
197 #define PLATFORM_LPAR		0x0001
198 #define PLATFORM_POWERMAC	0x0400
199 #define PLATFORM_GENERIC	0x0500
200 
201 static int __prombss of_platform;
202 
203 static char __prombss prom_cmd_line[COMMAND_LINE_SIZE];
204 
205 static unsigned long __prombss prom_memory_limit;
206 
207 static unsigned long __prombss alloc_top;
208 static unsigned long __prombss alloc_top_high;
209 static unsigned long __prombss alloc_bottom;
210 static unsigned long __prombss rmo_top;
211 static unsigned long __prombss ram_top;
212 
213 static struct mem_map_entry __prombss mem_reserve_map[MEM_RESERVE_MAP_SIZE];
214 static int __prombss mem_reserve_cnt;
215 
216 static cell_t __prombss regbuf[1024];
217 
218 static bool  __prombss rtas_has_query_cpu_stopped;
219 
220 
221 /*
222  * Error results ... some OF calls will return "-1" on error, some
223  * will return 0, some will return either. To simplify, here are
224  * macros to use with any ihandle or phandle return value to check if
225  * it is valid
226  */
227 
228 #define PROM_ERROR		(-1u)
229 #define PHANDLE_VALID(p)	((p) != 0 && (p) != PROM_ERROR)
230 #define IHANDLE_VALID(i)	((i) != 0 && (i) != PROM_ERROR)
231 
232 /* Copied from lib/string.c and lib/kstrtox.c */
233 
234 static int __init prom_strcmp(const char *cs, const char *ct)
235 {
236 	unsigned char c1, c2;
237 
238 	while (1) {
239 		c1 = *cs++;
240 		c2 = *ct++;
241 		if (c1 != c2)
242 			return c1 < c2 ? -1 : 1;
243 		if (!c1)
244 			break;
245 	}
246 	return 0;
247 }
248 
249 static ssize_t __init prom_strscpy_pad(char *dest, const char *src, size_t n)
250 {
251 	ssize_t rc;
252 	size_t i;
253 
254 	if (n == 0 || n > INT_MAX)
255 		return -E2BIG;
256 
257 	// Copy up to n bytes
258 	for (i = 0; i < n && src[i] != '\0'; i++)
259 		dest[i] = src[i];
260 
261 	rc = i;
262 
263 	// If we copied all n then we have run out of space for the nul
264 	if (rc == n) {
265 		// Rewind by one character to ensure nul termination
266 		i--;
267 		rc = -E2BIG;
268 	}
269 
270 	for (; i < n; i++)
271 		dest[i] = '\0';
272 
273 	return rc;
274 }
275 
276 static int __init prom_strncmp(const char *cs, const char *ct, size_t count)
277 {
278 	unsigned char c1, c2;
279 
280 	while (count) {
281 		c1 = *cs++;
282 		c2 = *ct++;
283 		if (c1 != c2)
284 			return c1 < c2 ? -1 : 1;
285 		if (!c1)
286 			break;
287 		count--;
288 	}
289 	return 0;
290 }
291 
292 static size_t __init prom_strlen(const char *s)
293 {
294 	const char *sc;
295 
296 	for (sc = s; *sc != '\0'; ++sc)
297 		/* nothing */;
298 	return sc - s;
299 }
300 
301 static int __init prom_memcmp(const void *cs, const void *ct, size_t count)
302 {
303 	const unsigned char *su1, *su2;
304 	int res = 0;
305 
306 	for (su1 = cs, su2 = ct; 0 < count; ++su1, ++su2, count--)
307 		if ((res = *su1 - *su2) != 0)
308 			break;
309 	return res;
310 }
311 
312 static char __init *prom_strstr(const char *s1, const char *s2)
313 {
314 	size_t l1, l2;
315 
316 	l2 = prom_strlen(s2);
317 	if (!l2)
318 		return (char *)s1;
319 	l1 = prom_strlen(s1);
320 	while (l1 >= l2) {
321 		l1--;
322 		if (!prom_memcmp(s1, s2, l2))
323 			return (char *)s1;
324 		s1++;
325 	}
326 	return NULL;
327 }
328 
329 static size_t __init prom_strlcat(char *dest, const char *src, size_t count)
330 {
331 	size_t dsize = prom_strlen(dest);
332 	size_t len = prom_strlen(src);
333 	size_t res = dsize + len;
334 
335 	/* This would be a bug */
336 	if (dsize >= count)
337 		return count;
338 
339 	dest += dsize;
340 	count -= dsize;
341 	if (len >= count)
342 		len = count-1;
343 	memcpy(dest, src, len);
344 	dest[len] = 0;
345 	return res;
346 
347 }
348 
349 #ifdef CONFIG_PPC_PSERIES
350 static int __init prom_strtobool(const char *s, bool *res)
351 {
352 	if (!s)
353 		return -EINVAL;
354 
355 	switch (s[0]) {
356 	case 'y':
357 	case 'Y':
358 	case '1':
359 		*res = true;
360 		return 0;
361 	case 'n':
362 	case 'N':
363 	case '0':
364 		*res = false;
365 		return 0;
366 	case 'o':
367 	case 'O':
368 		switch (s[1]) {
369 		case 'n':
370 		case 'N':
371 			*res = true;
372 			return 0;
373 		case 'f':
374 		case 'F':
375 			*res = false;
376 			return 0;
377 		default:
378 			break;
379 		}
380 		break;
381 	default:
382 		break;
383 	}
384 
385 	return -EINVAL;
386 }
387 #endif
388 
389 /* This is the one and *ONLY* place where we actually call open
390  * firmware.
391  */
392 
393 static int __init call_prom(const char *service, int nargs, int nret, ...)
394 {
395 	int i;
396 	struct prom_args args;
397 	va_list list;
398 
399 	args.service = cpu_to_be32(ADDR(service));
400 	args.nargs = cpu_to_be32(nargs);
401 	args.nret = cpu_to_be32(nret);
402 
403 	va_start(list, nret);
404 	for (i = 0; i < nargs; i++)
405 		args.args[i] = cpu_to_be32(va_arg(list, prom_arg_t));
406 	va_end(list);
407 
408 	for (i = 0; i < nret; i++)
409 		args.args[nargs+i] = 0;
410 
411 	if (enter_prom(&args, prom_entry) < 0)
412 		return PROM_ERROR;
413 
414 	return (nret > 0) ? be32_to_cpu(args.args[nargs]) : 0;
415 }
416 
417 static int __init call_prom_ret(const char *service, int nargs, int nret,
418 				prom_arg_t *rets, ...)
419 {
420 	int i;
421 	struct prom_args args;
422 	va_list list;
423 
424 	args.service = cpu_to_be32(ADDR(service));
425 	args.nargs = cpu_to_be32(nargs);
426 	args.nret = cpu_to_be32(nret);
427 
428 	va_start(list, rets);
429 	for (i = 0; i < nargs; i++)
430 		args.args[i] = cpu_to_be32(va_arg(list, prom_arg_t));
431 	va_end(list);
432 
433 	for (i = 0; i < nret; i++)
434 		args.args[nargs+i] = 0;
435 
436 	if (enter_prom(&args, prom_entry) < 0)
437 		return PROM_ERROR;
438 
439 	if (rets != NULL)
440 		for (i = 1; i < nret; ++i)
441 			rets[i-1] = be32_to_cpu(args.args[nargs+i]);
442 
443 	return (nret > 0) ? be32_to_cpu(args.args[nargs]) : 0;
444 }
445 
446 
447 static void __init prom_print(const char *msg)
448 {
449 	const char *p, *q;
450 
451 	if (prom.stdout == 0)
452 		return;
453 
454 	for (p = msg; *p != 0; p = q) {
455 		for (q = p; *q != 0 && *q != '\n'; ++q)
456 			;
457 		if (q > p)
458 			call_prom("write", 3, 1, prom.stdout, p, q - p);
459 		if (*q == 0)
460 			break;
461 		++q;
462 		call_prom("write", 3, 1, prom.stdout, ADDR("\r\n"), 2);
463 	}
464 }
465 
466 
467 /*
468  * Both prom_print_hex & prom_print_dec takes an unsigned long as input so that
469  * we do not need __udivdi3 or __umoddi3 on 32bits.
470  */
471 static void __init prom_print_hex(unsigned long val)
472 {
473 	int i, nibbles = sizeof(val)*2;
474 	char buf[sizeof(val)*2+1];
475 
476 	for (i = nibbles-1;  i >= 0;  i--) {
477 		buf[i] = (val & 0xf) + '0';
478 		if (buf[i] > '9')
479 			buf[i] += ('a'-'0'-10);
480 		val >>= 4;
481 	}
482 	buf[nibbles] = '\0';
483 	call_prom("write", 3, 1, prom.stdout, buf, nibbles);
484 }
485 
486 /* max number of decimal digits in an unsigned long */
487 #define UL_DIGITS 21
488 static void __init prom_print_dec(unsigned long val)
489 {
490 	int i, size;
491 	char buf[UL_DIGITS+1];
492 
493 	for (i = UL_DIGITS-1; i >= 0;  i--) {
494 		buf[i] = (val % 10) + '0';
495 		val = val/10;
496 		if (val == 0)
497 			break;
498 	}
499 	/* shift stuff down */
500 	size = UL_DIGITS - i;
501 	call_prom("write", 3, 1, prom.stdout, buf+i, size);
502 }
503 
504 __printf(1, 2)
505 static void __init prom_printf(const char *format, ...)
506 {
507 	const char *p, *q, *s;
508 	va_list args;
509 	unsigned long v;
510 	long vs;
511 	int n = 0;
512 
513 	va_start(args, format);
514 	for (p = format; *p != 0; p = q) {
515 		for (q = p; *q != 0 && *q != '\n' && *q != '%'; ++q)
516 			;
517 		if (q > p)
518 			call_prom("write", 3, 1, prom.stdout, p, q - p);
519 		if (*q == 0)
520 			break;
521 		if (*q == '\n') {
522 			++q;
523 			call_prom("write", 3, 1, prom.stdout,
524 				  ADDR("\r\n"), 2);
525 			continue;
526 		}
527 		++q;
528 		if (*q == 0)
529 			break;
530 		while (*q == 'l') {
531 			++q;
532 			++n;
533 		}
534 		switch (*q) {
535 		case 's':
536 			++q;
537 			s = va_arg(args, const char *);
538 			prom_print(s);
539 			break;
540 		case 'x':
541 			++q;
542 			switch (n) {
543 			case 0:
544 				v = va_arg(args, unsigned int);
545 				break;
546 			case 1:
547 				v = va_arg(args, unsigned long);
548 				break;
549 			case 2:
550 			default:
551 				v = va_arg(args, unsigned long long);
552 				break;
553 			}
554 			prom_print_hex(v);
555 			break;
556 		case 'u':
557 			++q;
558 			switch (n) {
559 			case 0:
560 				v = va_arg(args, unsigned int);
561 				break;
562 			case 1:
563 				v = va_arg(args, unsigned long);
564 				break;
565 			case 2:
566 			default:
567 				v = va_arg(args, unsigned long long);
568 				break;
569 			}
570 			prom_print_dec(v);
571 			break;
572 		case 'd':
573 			++q;
574 			switch (n) {
575 			case 0:
576 				vs = va_arg(args, int);
577 				break;
578 			case 1:
579 				vs = va_arg(args, long);
580 				break;
581 			case 2:
582 			default:
583 				vs = va_arg(args, long long);
584 				break;
585 			}
586 			if (vs < 0) {
587 				prom_print("-");
588 				vs = -vs;
589 			}
590 			prom_print_dec(vs);
591 			break;
592 		}
593 	}
594 	va_end(args);
595 }
596 
597 
598 static unsigned int __init prom_claim(unsigned long virt, unsigned long size,
599 				unsigned long align)
600 {
601 
602 	if (align == 0 && (OF_WORKAROUNDS & OF_WA_CLAIM)) {
603 		/*
604 		 * Old OF requires we claim physical and virtual separately
605 		 * and then map explicitly (assuming virtual mode)
606 		 */
607 		int ret;
608 		prom_arg_t result;
609 
610 		ret = call_prom_ret("call-method", 5, 2, &result,
611 				    ADDR("claim"), prom.memory,
612 				    align, size, virt);
613 		if (ret != 0 || result == -1)
614 			return -1;
615 		ret = call_prom_ret("call-method", 5, 2, &result,
616 				    ADDR("claim"), prom.mmumap,
617 				    align, size, virt);
618 		if (ret != 0) {
619 			call_prom("call-method", 4, 1, ADDR("release"),
620 				  prom.memory, size, virt);
621 			return -1;
622 		}
623 		/* the 0x12 is M (coherence) + PP == read/write */
624 		call_prom("call-method", 6, 1,
625 			  ADDR("map"), prom.mmumap, 0x12, size, virt, virt);
626 		return virt;
627 	}
628 	return call_prom("claim", 3, 1, (prom_arg_t)virt, (prom_arg_t)size,
629 			 (prom_arg_t)align);
630 }
631 
632 static void __init __attribute__((noreturn)) prom_panic(const char *reason)
633 {
634 	prom_print(reason);
635 	/* Do not call exit because it clears the screen on pmac
636 	 * it also causes some sort of double-fault on early pmacs */
637 	if (of_platform == PLATFORM_POWERMAC)
638 		asm("trap\n");
639 
640 	/* ToDo: should put up an SRC here on pSeries */
641 	call_prom("exit", 0, 0);
642 
643 	for (;;)			/* should never get here */
644 		;
645 }
646 
647 
648 static int __init prom_next_node(phandle *nodep)
649 {
650 	phandle node;
651 
652 	if ((node = *nodep) != 0
653 	    && (*nodep = call_prom("child", 1, 1, node)) != 0)
654 		return 1;
655 	if ((*nodep = call_prom("peer", 1, 1, node)) != 0)
656 		return 1;
657 	for (;;) {
658 		if ((node = call_prom("parent", 1, 1, node)) == 0)
659 			return 0;
660 		if ((*nodep = call_prom("peer", 1, 1, node)) != 0)
661 			return 1;
662 	}
663 }
664 
665 static inline int __init prom_getprop(phandle node, const char *pname,
666 				      void *value, size_t valuelen)
667 {
668 	return call_prom("getprop", 4, 1, node, ADDR(pname),
669 			 (u32)(unsigned long) value, (u32) valuelen);
670 }
671 
672 static inline int __init prom_getproplen(phandle node, const char *pname)
673 {
674 	return call_prom("getproplen", 2, 1, node, ADDR(pname));
675 }
676 
677 static void __init add_string(char **str, const char *q)
678 {
679 	char *p = *str;
680 
681 	while (*q)
682 		*p++ = *q++;
683 	*p++ = ' ';
684 	*str = p;
685 }
686 
687 static char *__init tohex(unsigned int x)
688 {
689 	static const char digits[] __initconst = "0123456789abcdef";
690 	static char result[9] __prombss;
691 	int i;
692 
693 	result[8] = 0;
694 	i = 8;
695 	do {
696 		--i;
697 		result[i] = digits[x & 0xf];
698 		x >>= 4;
699 	} while (x != 0 && i > 0);
700 	return &result[i];
701 }
702 
703 static int __init prom_setprop(phandle node, const char *nodename,
704 			       const char *pname, void *value, size_t valuelen)
705 {
706 	char cmd[256], *p;
707 
708 	if (!(OF_WORKAROUNDS & OF_WA_LONGTRAIL))
709 		return call_prom("setprop", 4, 1, node, ADDR(pname),
710 				 (u32)(unsigned long) value, (u32) valuelen);
711 
712 	/* gah... setprop doesn't work on longtrail, have to use interpret */
713 	p = cmd;
714 	add_string(&p, "dev");
715 	add_string(&p, nodename);
716 	add_string(&p, tohex((u32)(unsigned long) value));
717 	add_string(&p, tohex(valuelen));
718 	add_string(&p, tohex(ADDR(pname)));
719 	add_string(&p, tohex(prom_strlen(pname)));
720 	add_string(&p, "property");
721 	*p = 0;
722 	return call_prom("interpret", 1, 1, (u32)(unsigned long) cmd);
723 }
724 
725 /* We can't use the standard versions because of relocation headaches. */
726 #define prom_isxdigit(c) \
727 	(('0' <= (c) && (c) <= '9') || ('a' <= (c) && (c) <= 'f') || ('A' <= (c) && (c) <= 'F'))
728 
729 #define prom_isdigit(c)	('0' <= (c) && (c) <= '9')
730 #define prom_islower(c)	('a' <= (c) && (c) <= 'z')
731 #define prom_toupper(c)	(prom_islower(c) ? ((c) - 'a' + 'A') : (c))
732 
733 static unsigned long __init prom_strtoul(const char *cp, const char **endp)
734 {
735 	unsigned long result = 0, base = 10, value;
736 
737 	if (*cp == '0') {
738 		base = 8;
739 		cp++;
740 		if (prom_toupper(*cp) == 'X') {
741 			cp++;
742 			base = 16;
743 		}
744 	}
745 
746 	while (prom_isxdigit(*cp) &&
747 	       (value = prom_isdigit(*cp) ? *cp - '0' : prom_toupper(*cp) - 'A' + 10) < base) {
748 		result = result * base + value;
749 		cp++;
750 	}
751 
752 	if (endp)
753 		*endp = cp;
754 
755 	return result;
756 }
757 
758 static unsigned long __init prom_memparse(const char *ptr, const char **retptr)
759 {
760 	unsigned long ret = prom_strtoul(ptr, retptr);
761 	int shift = 0;
762 
763 	/*
764 	 * We can't use a switch here because GCC *may* generate a
765 	 * jump table which won't work, because we're not running at
766 	 * the address we're linked at.
767 	 */
768 	if ('G' == **retptr || 'g' == **retptr)
769 		shift = 30;
770 
771 	if ('M' == **retptr || 'm' == **retptr)
772 		shift = 20;
773 
774 	if ('K' == **retptr || 'k' == **retptr)
775 		shift = 10;
776 
777 	if (shift) {
778 		ret <<= shift;
779 		(*retptr)++;
780 	}
781 
782 	return ret;
783 }
784 
785 /*
786  * Early parsing of the command line passed to the kernel, used for
787  * "mem=x" and the options that affect the iommu
788  */
789 static void __init early_cmdline_parse(void)
790 {
791 	const char *opt;
792 
793 	char *p;
794 	int l = 0;
795 
796 	prom_cmd_line[0] = 0;
797 	p = prom_cmd_line;
798 
799 	if (!IS_ENABLED(CONFIG_CMDLINE_FORCE) && (long)prom.chosen > 0)
800 		l = prom_getprop(prom.chosen, "bootargs", p, COMMAND_LINE_SIZE-1);
801 
802 	if (IS_ENABLED(CONFIG_CMDLINE_EXTEND) || l <= 0 || p[0] == '\0')
803 		prom_strlcat(prom_cmd_line, " " CONFIG_CMDLINE,
804 			     sizeof(prom_cmd_line));
805 
806 	prom_printf("command line: %s\n", prom_cmd_line);
807 
808 #ifdef CONFIG_PPC64
809 	opt = prom_strstr(prom_cmd_line, "iommu=");
810 	if (opt) {
811 		prom_printf("iommu opt is: %s\n", opt);
812 		opt += 6;
813 		while (*opt && *opt == ' ')
814 			opt++;
815 		if (!prom_strncmp(opt, "off", 3))
816 			prom_iommu_off = 1;
817 		else if (!prom_strncmp(opt, "force", 5))
818 			prom_iommu_force_on = 1;
819 	}
820 #endif
821 	opt = prom_strstr(prom_cmd_line, "mem=");
822 	if (opt) {
823 		opt += 4;
824 		prom_memory_limit = prom_memparse(opt, (const char **)&opt);
825 #ifdef CONFIG_PPC64
826 		/* Align to 16 MB == size of ppc64 large page */
827 		prom_memory_limit = ALIGN(prom_memory_limit, 0x1000000);
828 #endif
829 	}
830 
831 #ifdef CONFIG_PPC_PSERIES
832 	prom_radix_disable = !IS_ENABLED(CONFIG_PPC_RADIX_MMU_DEFAULT);
833 	opt = prom_strstr(prom_cmd_line, "disable_radix");
834 	if (opt) {
835 		opt += 13;
836 		if (*opt && *opt == '=') {
837 			bool val;
838 
839 			if (prom_strtobool(++opt, &val))
840 				prom_radix_disable = false;
841 			else
842 				prom_radix_disable = val;
843 		} else
844 			prom_radix_disable = true;
845 	}
846 	if (prom_radix_disable)
847 		prom_debug("Radix disabled from cmdline\n");
848 
849 	opt = prom_strstr(prom_cmd_line, "radix_hcall_invalidate=on");
850 	if (opt) {
851 		prom_radix_gtse_disable = true;
852 		prom_debug("Radix GTSE disabled from cmdline\n");
853 	}
854 
855 	opt = prom_strstr(prom_cmd_line, "xive=off");
856 	if (opt) {
857 		prom_xive_disable = true;
858 		prom_debug("XIVE disabled from cmdline\n");
859 	}
860 #endif /* CONFIG_PPC_PSERIES */
861 
862 #ifdef CONFIG_PPC_SVM
863 	opt = prom_strstr(prom_cmd_line, "svm=");
864 	if (opt) {
865 		bool val;
866 
867 		opt += sizeof("svm=") - 1;
868 		if (!prom_strtobool(opt, &val))
869 			prom_svm_enable = val;
870 	}
871 #endif /* CONFIG_PPC_SVM */
872 }
873 
874 #ifdef CONFIG_PPC_PSERIES
875 /*
876  * The architecture vector has an array of PVR mask/value pairs,
877  * followed by # option vectors - 1, followed by the option vectors.
878  *
879  * See prom.h for the definition of the bits specified in the
880  * architecture vector.
881  */
882 
883 /* Firmware expects the value to be n - 1, where n is the # of vectors */
884 #define NUM_VECTORS(n)		((n) - 1)
885 
886 /*
887  * Firmware expects 1 + n - 2, where n is the length of the option vector in
888  * bytes. The 1 accounts for the length byte itself, the - 2 .. ?
889  */
890 #define VECTOR_LENGTH(n)	(1 + (n) - 2)
891 
892 struct option_vector1 {
893 	u8 byte1;
894 	u8 arch_versions;
895 	u8 arch_versions3;
896 } __packed;
897 
898 struct option_vector2 {
899 	u8 byte1;
900 	__be16 reserved;
901 	__be32 real_base;
902 	__be32 real_size;
903 	__be32 virt_base;
904 	__be32 virt_size;
905 	__be32 load_base;
906 	__be32 min_rma;
907 	__be32 min_load;
908 	u8 min_rma_percent;
909 	u8 max_pft_size;
910 } __packed;
911 
912 struct option_vector3 {
913 	u8 byte1;
914 	u8 byte2;
915 } __packed;
916 
917 struct option_vector4 {
918 	u8 byte1;
919 	u8 min_vp_cap;
920 } __packed;
921 
922 struct option_vector5 {
923 	u8 byte1;
924 	u8 byte2;
925 	u8 byte3;
926 	u8 cmo;
927 	u8 associativity;
928 	u8 bin_opts;
929 	u8 micro_checkpoint;
930 	u8 reserved0;
931 	__be32 max_cpus;
932 	__be16 papr_level;
933 	__be16 reserved1;
934 	u8 platform_facilities;
935 	u8 reserved2;
936 	__be16 reserved3;
937 	u8 subprocessors;
938 	u8 byte22;
939 	u8 intarch;
940 	u8 mmu;
941 	u8 hash_ext;
942 	u8 radix_ext;
943 } __packed;
944 
945 struct option_vector6 {
946 	u8 reserved;
947 	u8 secondary_pteg;
948 	u8 os_name;
949 } __packed;
950 
951 struct option_vector7 {
952 	u8 os_id[256];
953 } __packed;
954 
955 struct ibm_arch_vec {
956 	struct { u32 mask, val; } pvrs[14];
957 
958 	u8 num_vectors;
959 
960 	u8 vec1_len;
961 	struct option_vector1 vec1;
962 
963 	u8 vec2_len;
964 	struct option_vector2 vec2;
965 
966 	u8 vec3_len;
967 	struct option_vector3 vec3;
968 
969 	u8 vec4_len;
970 	struct option_vector4 vec4;
971 
972 	u8 vec5_len;
973 	struct option_vector5 vec5;
974 
975 	u8 vec6_len;
976 	struct option_vector6 vec6;
977 
978 	u8 vec7_len;
979 	struct option_vector7 vec7;
980 } __packed;
981 
982 static const struct ibm_arch_vec ibm_architecture_vec_template __initconst = {
983 	.pvrs = {
984 		{
985 			.mask = cpu_to_be32(0xfffe0000), /* POWER5/POWER5+ */
986 			.val  = cpu_to_be32(0x003a0000),
987 		},
988 		{
989 			.mask = cpu_to_be32(0xffff0000), /* POWER6 */
990 			.val  = cpu_to_be32(0x003e0000),
991 		},
992 		{
993 			.mask = cpu_to_be32(0xffff0000), /* POWER7 */
994 			.val  = cpu_to_be32(0x003f0000),
995 		},
996 		{
997 			.mask = cpu_to_be32(0xffff0000), /* POWER8E */
998 			.val  = cpu_to_be32(0x004b0000),
999 		},
1000 		{
1001 			.mask = cpu_to_be32(0xffff0000), /* POWER8NVL */
1002 			.val  = cpu_to_be32(0x004c0000),
1003 		},
1004 		{
1005 			.mask = cpu_to_be32(0xffff0000), /* POWER8 */
1006 			.val  = cpu_to_be32(0x004d0000),
1007 		},
1008 		{
1009 			.mask = cpu_to_be32(0xffff0000), /* POWER9 */
1010 			.val  = cpu_to_be32(0x004e0000),
1011 		},
1012 		{
1013 			.mask = cpu_to_be32(0xffff0000), /* POWER10 */
1014 			.val  = cpu_to_be32(0x00800000),
1015 		},
1016 		{
1017 			.mask = cpu_to_be32(0xffffffff), /* all 3.1-compliant */
1018 			.val  = cpu_to_be32(0x0f000006),
1019 		},
1020 		{
1021 			.mask = cpu_to_be32(0xffffffff), /* all 3.00-compliant */
1022 			.val  = cpu_to_be32(0x0f000005),
1023 		},
1024 		{
1025 			.mask = cpu_to_be32(0xffffffff), /* all 2.07-compliant */
1026 			.val  = cpu_to_be32(0x0f000004),
1027 		},
1028 		{
1029 			.mask = cpu_to_be32(0xffffffff), /* all 2.06-compliant */
1030 			.val  = cpu_to_be32(0x0f000003),
1031 		},
1032 		{
1033 			.mask = cpu_to_be32(0xffffffff), /* all 2.05-compliant */
1034 			.val  = cpu_to_be32(0x0f000002),
1035 		},
1036 		{
1037 			.mask = cpu_to_be32(0xfffffffe), /* all 2.04-compliant and earlier */
1038 			.val  = cpu_to_be32(0x0f000001),
1039 		},
1040 	},
1041 
1042 	.num_vectors = NUM_VECTORS(6),
1043 
1044 	.vec1_len = VECTOR_LENGTH(sizeof(struct option_vector1)),
1045 	.vec1 = {
1046 		.byte1 = 0,
1047 		.arch_versions = OV1_PPC_2_00 | OV1_PPC_2_01 | OV1_PPC_2_02 | OV1_PPC_2_03 |
1048 				 OV1_PPC_2_04 | OV1_PPC_2_05 | OV1_PPC_2_06 | OV1_PPC_2_07,
1049 		.arch_versions3 = OV1_PPC_3_00 | OV1_PPC_3_1,
1050 	},
1051 
1052 	.vec2_len = VECTOR_LENGTH(sizeof(struct option_vector2)),
1053 	/* option vector 2: Open Firmware options supported */
1054 	.vec2 = {
1055 		.byte1 = OV2_REAL_MODE,
1056 		.reserved = 0,
1057 		.real_base = cpu_to_be32(0xffffffff),
1058 		.real_size = cpu_to_be32(0xffffffff),
1059 		.virt_base = cpu_to_be32(0xffffffff),
1060 		.virt_size = cpu_to_be32(0xffffffff),
1061 		.load_base = cpu_to_be32(0xffffffff),
1062 		.min_rma = cpu_to_be32(512),		/* 512MB min RMA */
1063 		.min_load = cpu_to_be32(0xffffffff),	/* full client load */
1064 		.min_rma_percent = 0,	/* min RMA percentage of total RAM */
1065 		.max_pft_size = 48,	/* max log_2(hash table size) */
1066 	},
1067 
1068 	.vec3_len = VECTOR_LENGTH(sizeof(struct option_vector3)),
1069 	/* option vector 3: processor options supported */
1070 	.vec3 = {
1071 		.byte1 = 0,			/* don't ignore, don't halt */
1072 		.byte2 = OV3_FP | OV3_VMX | OV3_DFP,
1073 	},
1074 
1075 	.vec4_len = VECTOR_LENGTH(sizeof(struct option_vector4)),
1076 	/* option vector 4: IBM PAPR implementation */
1077 	.vec4 = {
1078 		.byte1 = 0,			/* don't halt */
1079 		.min_vp_cap = OV4_MIN_ENT_CAP,	/* minimum VP entitled capacity */
1080 	},
1081 
1082 	.vec5_len = VECTOR_LENGTH(sizeof(struct option_vector5)),
1083 	/* option vector 5: PAPR/OF options */
1084 	.vec5 = {
1085 		.byte1 = 0,				/* don't ignore, don't halt */
1086 		.byte2 = OV5_FEAT(OV5_LPAR) | OV5_FEAT(OV5_SPLPAR) | OV5_FEAT(OV5_LARGE_PAGES) |
1087 		OV5_FEAT(OV5_DRCONF_MEMORY) | OV5_FEAT(OV5_DONATE_DEDICATE_CPU) |
1088 #ifdef CONFIG_PCI_MSI
1089 		/* PCIe/MSI support.  Without MSI full PCIe is not supported */
1090 		OV5_FEAT(OV5_MSI),
1091 #else
1092 		0,
1093 #endif
1094 		.byte3 = 0,
1095 		.cmo =
1096 #ifdef CONFIG_PPC_SMLPAR
1097 		OV5_FEAT(OV5_CMO) | OV5_FEAT(OV5_XCMO),
1098 #else
1099 		0,
1100 #endif
1101 		.associativity = OV5_FEAT(OV5_FORM1_AFFINITY) | OV5_FEAT(OV5_PRRN) |
1102 		OV5_FEAT(OV5_FORM2_AFFINITY),
1103 		.bin_opts = OV5_FEAT(OV5_RESIZE_HPT) | OV5_FEAT(OV5_HP_EVT),
1104 		.micro_checkpoint = 0,
1105 		.reserved0 = 0,
1106 		.max_cpus = cpu_to_be32(NR_CPUS),	/* number of cores supported */
1107 		.papr_level = 0,
1108 		.reserved1 = 0,
1109 		.platform_facilities = OV5_FEAT(OV5_PFO_HW_RNG) | OV5_FEAT(OV5_PFO_HW_ENCR) | OV5_FEAT(OV5_PFO_HW_842),
1110 		.reserved2 = 0,
1111 		.reserved3 = 0,
1112 		.subprocessors = 1,
1113 		.byte22 = OV5_FEAT(OV5_DRMEM_V2) | OV5_FEAT(OV5_DRC_INFO),
1114 		.intarch = 0,
1115 		.mmu = 0,
1116 		.hash_ext = 0,
1117 		.radix_ext = 0,
1118 	},
1119 
1120 	/* option vector 6: IBM PAPR hints */
1121 	.vec6_len = VECTOR_LENGTH(sizeof(struct option_vector6)),
1122 	.vec6 = {
1123 		.reserved = 0,
1124 		.secondary_pteg = 0,
1125 		.os_name = OV6_LINUX,
1126 	},
1127 
1128 	/* option vector 7: OS Identification */
1129 	.vec7_len = VECTOR_LENGTH(sizeof(struct option_vector7)),
1130 };
1131 
1132 static struct ibm_arch_vec __prombss ibm_architecture_vec  ____cacheline_aligned;
1133 
1134 /* Old method - ELF header with PT_NOTE sections only works on BE */
1135 #ifdef __BIG_ENDIAN__
1136 static const struct fake_elf {
1137 	Elf32_Ehdr	elfhdr;
1138 	Elf32_Phdr	phdr[2];
1139 	struct chrpnote {
1140 		u32	namesz;
1141 		u32	descsz;
1142 		u32	type;
1143 		char	name[8];	/* "PowerPC" */
1144 		struct chrpdesc {
1145 			u32	real_mode;
1146 			u32	real_base;
1147 			u32	real_size;
1148 			u32	virt_base;
1149 			u32	virt_size;
1150 			u32	load_base;
1151 		} chrpdesc;
1152 	} chrpnote;
1153 	struct rpanote {
1154 		u32	namesz;
1155 		u32	descsz;
1156 		u32	type;
1157 		char	name[24];	/* "IBM,RPA-Client-Config" */
1158 		struct rpadesc {
1159 			u32	lpar_affinity;
1160 			u32	min_rmo_size;
1161 			u32	min_rmo_percent;
1162 			u32	max_pft_size;
1163 			u32	splpar;
1164 			u32	min_load;
1165 			u32	new_mem_def;
1166 			u32	ignore_me;
1167 		} rpadesc;
1168 	} rpanote;
1169 } fake_elf __initconst = {
1170 	.elfhdr = {
1171 		.e_ident = { 0x7f, 'E', 'L', 'F',
1172 			     ELFCLASS32, ELFDATA2MSB, EV_CURRENT },
1173 		.e_type = ET_EXEC,	/* yeah right */
1174 		.e_machine = EM_PPC,
1175 		.e_version = EV_CURRENT,
1176 		.e_phoff = offsetof(struct fake_elf, phdr),
1177 		.e_phentsize = sizeof(Elf32_Phdr),
1178 		.e_phnum = 2
1179 	},
1180 	.phdr = {
1181 		[0] = {
1182 			.p_type = PT_NOTE,
1183 			.p_offset = offsetof(struct fake_elf, chrpnote),
1184 			.p_filesz = sizeof(struct chrpnote)
1185 		}, [1] = {
1186 			.p_type = PT_NOTE,
1187 			.p_offset = offsetof(struct fake_elf, rpanote),
1188 			.p_filesz = sizeof(struct rpanote)
1189 		}
1190 	},
1191 	.chrpnote = {
1192 		.namesz = sizeof("PowerPC"),
1193 		.descsz = sizeof(struct chrpdesc),
1194 		.type = 0x1275,
1195 		.name = "PowerPC",
1196 		.chrpdesc = {
1197 			.real_mode = ~0U,	/* ~0 means "don't care" */
1198 			.real_base = ~0U,
1199 			.real_size = ~0U,
1200 			.virt_base = ~0U,
1201 			.virt_size = ~0U,
1202 			.load_base = ~0U
1203 		},
1204 	},
1205 	.rpanote = {
1206 		.namesz = sizeof("IBM,RPA-Client-Config"),
1207 		.descsz = sizeof(struct rpadesc),
1208 		.type = 0x12759999,
1209 		.name = "IBM,RPA-Client-Config",
1210 		.rpadesc = {
1211 			.lpar_affinity = 0,
1212 			.min_rmo_size = 64,	/* in megabytes */
1213 			.min_rmo_percent = 0,
1214 			.max_pft_size = 48,	/* 2^48 bytes max PFT size */
1215 			.splpar = 1,
1216 			.min_load = ~0U,
1217 			.new_mem_def = 0
1218 		}
1219 	}
1220 };
1221 #endif /* __BIG_ENDIAN__ */
1222 
1223 static int __init prom_count_smt_threads(void)
1224 {
1225 	phandle node;
1226 	char type[64];
1227 	unsigned int plen;
1228 
1229 	/* Pick up th first CPU node we can find */
1230 	for (node = 0; prom_next_node(&node); ) {
1231 		type[0] = 0;
1232 		prom_getprop(node, "device_type", type, sizeof(type));
1233 
1234 		if (prom_strcmp(type, "cpu"))
1235 			continue;
1236 		/*
1237 		 * There is an entry for each smt thread, each entry being
1238 		 * 4 bytes long.  All cpus should have the same number of
1239 		 * smt threads, so return after finding the first.
1240 		 */
1241 		plen = prom_getproplen(node, "ibm,ppc-interrupt-server#s");
1242 		if (plen == PROM_ERROR)
1243 			break;
1244 		plen >>= 2;
1245 		prom_debug("Found %lu smt threads per core\n", (unsigned long)plen);
1246 
1247 		/* Sanity check */
1248 		if (plen < 1 || plen > 64) {
1249 			prom_printf("Threads per core %lu out of bounds, assuming 1\n",
1250 				    (unsigned long)plen);
1251 			return 1;
1252 		}
1253 		return plen;
1254 	}
1255 	prom_debug("No threads found, assuming 1 per core\n");
1256 
1257 	return 1;
1258 
1259 }
1260 
1261 static void __init prom_parse_mmu_model(u8 val,
1262 					struct platform_support *support)
1263 {
1264 	switch (val) {
1265 	case OV5_FEAT(OV5_MMU_DYNAMIC):
1266 	case OV5_FEAT(OV5_MMU_EITHER): /* Either Available */
1267 		prom_debug("MMU - either supported\n");
1268 		support->radix_mmu = !prom_radix_disable;
1269 		support->hash_mmu = true;
1270 		break;
1271 	case OV5_FEAT(OV5_MMU_RADIX): /* Only Radix */
1272 		prom_debug("MMU - radix only\n");
1273 		if (prom_radix_disable) {
1274 			/*
1275 			 * If we __have__ to do radix, we're better off ignoring
1276 			 * the command line rather than not booting.
1277 			 */
1278 			prom_printf("WARNING: Ignoring cmdline option disable_radix\n");
1279 		}
1280 		support->radix_mmu = true;
1281 		break;
1282 	case OV5_FEAT(OV5_MMU_HASH):
1283 		prom_debug("MMU - hash only\n");
1284 		support->hash_mmu = true;
1285 		break;
1286 	default:
1287 		prom_debug("Unknown mmu support option: 0x%x\n", val);
1288 		break;
1289 	}
1290 }
1291 
1292 static void __init prom_parse_xive_model(u8 val,
1293 					 struct platform_support *support)
1294 {
1295 	switch (val) {
1296 	case OV5_FEAT(OV5_XIVE_EITHER): /* Either Available */
1297 		prom_debug("XIVE - either mode supported\n");
1298 		support->xive = !prom_xive_disable;
1299 		break;
1300 	case OV5_FEAT(OV5_XIVE_EXPLOIT): /* Only Exploitation mode */
1301 		prom_debug("XIVE - exploitation mode supported\n");
1302 		if (prom_xive_disable) {
1303 			/*
1304 			 * If we __have__ to do XIVE, we're better off ignoring
1305 			 * the command line rather than not booting.
1306 			 */
1307 			prom_printf("WARNING: Ignoring cmdline option xive=off\n");
1308 		}
1309 		support->xive = true;
1310 		break;
1311 	case OV5_FEAT(OV5_XIVE_LEGACY): /* Only Legacy mode */
1312 		prom_debug("XIVE - legacy mode supported\n");
1313 		break;
1314 	default:
1315 		prom_debug("Unknown xive support option: 0x%x\n", val);
1316 		break;
1317 	}
1318 }
1319 
1320 static void __init prom_parse_platform_support(u8 index, u8 val,
1321 					       struct platform_support *support)
1322 {
1323 	switch (index) {
1324 	case OV5_INDX(OV5_MMU_SUPPORT): /* MMU Model */
1325 		prom_parse_mmu_model(val & OV5_FEAT(OV5_MMU_SUPPORT), support);
1326 		break;
1327 	case OV5_INDX(OV5_RADIX_GTSE): /* Radix Extensions */
1328 		if (val & OV5_FEAT(OV5_RADIX_GTSE))
1329 			support->radix_gtse = !prom_radix_gtse_disable;
1330 		break;
1331 	case OV5_INDX(OV5_XIVE_SUPPORT): /* Interrupt mode */
1332 		prom_parse_xive_model(val & OV5_FEAT(OV5_XIVE_SUPPORT),
1333 				      support);
1334 		break;
1335 	}
1336 }
1337 
1338 static void __init prom_check_platform_support(void)
1339 {
1340 	struct platform_support supported = {
1341 		.hash_mmu = false,
1342 		.radix_mmu = false,
1343 		.radix_gtse = false,
1344 		.xive = false
1345 	};
1346 	int prop_len = prom_getproplen(prom.chosen,
1347 				       "ibm,arch-vec-5-platform-support");
1348 
1349 	/*
1350 	 * First copy the architecture vec template
1351 	 *
1352 	 * use memcpy() instead of *vec = *vec_template so that GCC replaces it
1353 	 * by __memcpy() when KASAN is active
1354 	 */
1355 	memcpy(&ibm_architecture_vec, &ibm_architecture_vec_template,
1356 	       sizeof(ibm_architecture_vec));
1357 
1358 	prom_strscpy_pad(ibm_architecture_vec.vec7.os_id, linux_banner, 256);
1359 
1360 	if (prop_len > 1) {
1361 		int i;
1362 		u8 vec[8];
1363 		prom_debug("Found ibm,arch-vec-5-platform-support, len: %d\n",
1364 			   prop_len);
1365 		if (prop_len > sizeof(vec))
1366 			prom_printf("WARNING: ibm,arch-vec-5-platform-support longer than expected (len: %d)\n",
1367 				    prop_len);
1368 		prom_getprop(prom.chosen, "ibm,arch-vec-5-platform-support", &vec, sizeof(vec));
1369 		for (i = 0; i < prop_len; i += 2) {
1370 			prom_debug("%d: index = 0x%x val = 0x%x\n", i / 2, vec[i], vec[i + 1]);
1371 			prom_parse_platform_support(vec[i], vec[i + 1], &supported);
1372 		}
1373 	}
1374 
1375 	if (supported.radix_mmu && IS_ENABLED(CONFIG_PPC_RADIX_MMU)) {
1376 		/* Radix preferred - Check if GTSE is also supported */
1377 		prom_debug("Asking for radix\n");
1378 		ibm_architecture_vec.vec5.mmu = OV5_FEAT(OV5_MMU_RADIX);
1379 		if (supported.radix_gtse)
1380 			ibm_architecture_vec.vec5.radix_ext =
1381 					OV5_FEAT(OV5_RADIX_GTSE);
1382 		else
1383 			prom_debug("Radix GTSE isn't supported\n");
1384 	} else if (supported.hash_mmu) {
1385 		/* Default to hash mmu (if we can) */
1386 		prom_debug("Asking for hash\n");
1387 		ibm_architecture_vec.vec5.mmu = OV5_FEAT(OV5_MMU_HASH);
1388 	} else {
1389 		/* We're probably on a legacy hypervisor */
1390 		prom_debug("Assuming legacy hash support\n");
1391 	}
1392 
1393 	if (supported.xive) {
1394 		prom_debug("Asking for XIVE\n");
1395 		ibm_architecture_vec.vec5.intarch = OV5_FEAT(OV5_XIVE_EXPLOIT);
1396 	}
1397 }
1398 
1399 static void __init prom_send_capabilities(void)
1400 {
1401 	ihandle root;
1402 	prom_arg_t ret;
1403 	u32 cores;
1404 
1405 	/* Check ibm,arch-vec-5-platform-support and fixup vec5 if required */
1406 	prom_check_platform_support();
1407 
1408 	root = call_prom("open", 1, 1, ADDR("/"));
1409 	if (root != 0) {
1410 		/* We need to tell the FW about the number of cores we support.
1411 		 *
1412 		 * To do that, we count the number of threads on the first core
1413 		 * (we assume this is the same for all cores) and use it to
1414 		 * divide NR_CPUS.
1415 		 */
1416 
1417 		cores = DIV_ROUND_UP(NR_CPUS, prom_count_smt_threads());
1418 		prom_printf("Max number of cores passed to firmware: %u (NR_CPUS = %d)\n",
1419 			    cores, NR_CPUS);
1420 
1421 		ibm_architecture_vec.vec5.max_cpus = cpu_to_be32(cores);
1422 
1423 		/* try calling the ibm,client-architecture-support method */
1424 		prom_printf("Calling ibm,client-architecture-support...");
1425 		if (call_prom_ret("call-method", 3, 2, &ret,
1426 				  ADDR("ibm,client-architecture-support"),
1427 				  root,
1428 				  ADDR(&ibm_architecture_vec)) == 0) {
1429 			/* the call exists... */
1430 			if (ret)
1431 				prom_printf("\nWARNING: ibm,client-architecture"
1432 					    "-support call FAILED!\n");
1433 			call_prom("close", 1, 0, root);
1434 			prom_printf(" done\n");
1435 			return;
1436 		}
1437 		call_prom("close", 1, 0, root);
1438 		prom_printf(" not implemented\n");
1439 	}
1440 
1441 #ifdef __BIG_ENDIAN__
1442 	{
1443 		ihandle elfloader;
1444 
1445 		/* no ibm,client-architecture-support call, try the old way */
1446 		elfloader = call_prom("open", 1, 1,
1447 				      ADDR("/packages/elf-loader"));
1448 		if (elfloader == 0) {
1449 			prom_printf("couldn't open /packages/elf-loader\n");
1450 			return;
1451 		}
1452 		call_prom("call-method", 3, 1, ADDR("process-elf-header"),
1453 			  elfloader, ADDR(&fake_elf));
1454 		call_prom("close", 1, 0, elfloader);
1455 	}
1456 #endif /* __BIG_ENDIAN__ */
1457 }
1458 #endif /* CONFIG_PPC_PSERIES */
1459 
1460 /*
1461  * Memory allocation strategy... our layout is normally:
1462  *
1463  *  at 14Mb or more we have vmlinux, then a gap and initrd.  In some
1464  *  rare cases, initrd might end up being before the kernel though.
1465  *  We assume this won't override the final kernel at 0, we have no
1466  *  provision to handle that in this version, but it should hopefully
1467  *  never happen.
1468  *
1469  *  alloc_top is set to the top of RMO, eventually shrink down if the
1470  *  TCEs overlap
1471  *
1472  *  alloc_bottom is set to the top of kernel/initrd
1473  *
1474  *  from there, allocations are done this way : rtas is allocated
1475  *  topmost, and the device-tree is allocated from the bottom. We try
1476  *  to grow the device-tree allocation as we progress. If we can't,
1477  *  then we fail, we don't currently have a facility to restart
1478  *  elsewhere, but that shouldn't be necessary.
1479  *
1480  *  Note that calls to reserve_mem have to be done explicitly, memory
1481  *  allocated with either alloc_up or alloc_down isn't automatically
1482  *  reserved.
1483  */
1484 
1485 
1486 /*
1487  * Allocates memory in the RMO upward from the kernel/initrd
1488  *
1489  * When align is 0, this is a special case, it means to allocate in place
1490  * at the current location of alloc_bottom or fail (that is basically
1491  * extending the previous allocation). Used for the device-tree flattening
1492  */
1493 static unsigned long __init alloc_up(unsigned long size, unsigned long align)
1494 {
1495 	unsigned long base = alloc_bottom;
1496 	unsigned long addr = 0;
1497 
1498 	if (align)
1499 		base = ALIGN(base, align);
1500 	prom_debug("%s(%lx, %lx)\n", __func__, size, align);
1501 	if (ram_top == 0)
1502 		prom_panic("alloc_up() called with mem not initialized\n");
1503 
1504 	if (align)
1505 		base = ALIGN(alloc_bottom, align);
1506 	else
1507 		base = alloc_bottom;
1508 
1509 	for(; (base + size) <= alloc_top;
1510 	    base = ALIGN(base + 0x100000, align)) {
1511 		prom_debug("    trying: 0x%lx\n\r", base);
1512 		addr = (unsigned long)prom_claim(base, size, 0);
1513 		if (addr != PROM_ERROR && addr != 0)
1514 			break;
1515 		addr = 0;
1516 		if (align == 0)
1517 			break;
1518 	}
1519 	if (addr == 0)
1520 		return 0;
1521 	alloc_bottom = addr + size;
1522 
1523 	prom_debug(" -> %lx\n", addr);
1524 	prom_debug("  alloc_bottom : %lx\n", alloc_bottom);
1525 	prom_debug("  alloc_top    : %lx\n", alloc_top);
1526 	prom_debug("  alloc_top_hi : %lx\n", alloc_top_high);
1527 	prom_debug("  rmo_top      : %lx\n", rmo_top);
1528 	prom_debug("  ram_top      : %lx\n", ram_top);
1529 
1530 	return addr;
1531 }
1532 
1533 /*
1534  * Allocates memory downward, either from top of RMO, or if highmem
1535  * is set, from the top of RAM.  Note that this one doesn't handle
1536  * failures.  It does claim memory if highmem is not set.
1537  */
1538 static unsigned long __init alloc_down(unsigned long size, unsigned long align,
1539 				       int highmem)
1540 {
1541 	unsigned long base, addr = 0;
1542 
1543 	prom_debug("%s(%lx, %lx, %s)\n", __func__, size, align,
1544 		   highmem ? "(high)" : "(low)");
1545 	if (ram_top == 0)
1546 		prom_panic("alloc_down() called with mem not initialized\n");
1547 
1548 	if (highmem) {
1549 		/* Carve out storage for the TCE table. */
1550 		addr = ALIGN_DOWN(alloc_top_high - size, align);
1551 		if (addr <= alloc_bottom)
1552 			return 0;
1553 		/* Will we bump into the RMO ? If yes, check out that we
1554 		 * didn't overlap existing allocations there, if we did,
1555 		 * we are dead, we must be the first in town !
1556 		 */
1557 		if (addr < rmo_top) {
1558 			/* Good, we are first */
1559 			if (alloc_top == rmo_top)
1560 				alloc_top = rmo_top = addr;
1561 			else
1562 				return 0;
1563 		}
1564 		alloc_top_high = addr;
1565 		goto bail;
1566 	}
1567 
1568 	base = ALIGN_DOWN(alloc_top - size, align);
1569 	for (; base > alloc_bottom;
1570 	     base = ALIGN_DOWN(base - 0x100000, align))  {
1571 		prom_debug("    trying: 0x%lx\n\r", base);
1572 		addr = (unsigned long)prom_claim(base, size, 0);
1573 		if (addr != PROM_ERROR && addr != 0)
1574 			break;
1575 		addr = 0;
1576 	}
1577 	if (addr == 0)
1578 		return 0;
1579 	alloc_top = addr;
1580 
1581  bail:
1582 	prom_debug(" -> %lx\n", addr);
1583 	prom_debug("  alloc_bottom : %lx\n", alloc_bottom);
1584 	prom_debug("  alloc_top    : %lx\n", alloc_top);
1585 	prom_debug("  alloc_top_hi : %lx\n", alloc_top_high);
1586 	prom_debug("  rmo_top      : %lx\n", rmo_top);
1587 	prom_debug("  ram_top      : %lx\n", ram_top);
1588 
1589 	return addr;
1590 }
1591 
1592 /*
1593  * Parse a "reg" cell
1594  */
1595 static unsigned long __init prom_next_cell(int s, cell_t **cellp)
1596 {
1597 	cell_t *p = *cellp;
1598 	unsigned long r = 0;
1599 
1600 	/* Ignore more than 2 cells */
1601 	while (s > sizeof(unsigned long) / 4) {
1602 		p++;
1603 		s--;
1604 	}
1605 	r = be32_to_cpu(*p++);
1606 #ifdef CONFIG_PPC64
1607 	if (s > 1) {
1608 		r <<= 32;
1609 		r |= be32_to_cpu(*(p++));
1610 	}
1611 #endif
1612 	*cellp = p;
1613 	return r;
1614 }
1615 
1616 /*
1617  * Very dumb function for adding to the memory reserve list, but
1618  * we don't need anything smarter at this point
1619  *
1620  * XXX Eventually check for collisions.  They should NEVER happen.
1621  * If problems seem to show up, it would be a good start to track
1622  * them down.
1623  */
1624 static void __init reserve_mem(u64 base, u64 size)
1625 {
1626 	u64 top = base + size;
1627 	unsigned long cnt = mem_reserve_cnt;
1628 
1629 	if (size == 0)
1630 		return;
1631 
1632 	/* We need to always keep one empty entry so that we
1633 	 * have our terminator with "size" set to 0 since we are
1634 	 * dumb and just copy this entire array to the boot params
1635 	 */
1636 	base = ALIGN_DOWN(base, PAGE_SIZE);
1637 	top = ALIGN(top, PAGE_SIZE);
1638 	size = top - base;
1639 
1640 	if (cnt >= (MEM_RESERVE_MAP_SIZE - 1))
1641 		prom_panic("Memory reserve map exhausted !\n");
1642 	mem_reserve_map[cnt].base = cpu_to_be64(base);
1643 	mem_reserve_map[cnt].size = cpu_to_be64(size);
1644 	mem_reserve_cnt = cnt + 1;
1645 }
1646 
1647 /*
1648  * Initialize memory allocation mechanism, parse "memory" nodes and
1649  * obtain that way the top of memory and RMO to setup out local allocator
1650  */
1651 static void __init prom_init_mem(void)
1652 {
1653 	phandle node;
1654 	char type[64];
1655 	unsigned int plen;
1656 	cell_t *p, *endp;
1657 	__be32 val;
1658 	u32 rac, rsc;
1659 
1660 	/*
1661 	 * We iterate the memory nodes to find
1662 	 * 1) top of RMO (first node)
1663 	 * 2) top of memory
1664 	 */
1665 	val = cpu_to_be32(2);
1666 	prom_getprop(prom.root, "#address-cells", &val, sizeof(val));
1667 	rac = be32_to_cpu(val);
1668 	val = cpu_to_be32(1);
1669 	prom_getprop(prom.root, "#size-cells", &val, sizeof(rsc));
1670 	rsc = be32_to_cpu(val);
1671 	prom_debug("root_addr_cells: %x\n", rac);
1672 	prom_debug("root_size_cells: %x\n", rsc);
1673 
1674 	prom_debug("scanning memory:\n");
1675 
1676 	for (node = 0; prom_next_node(&node); ) {
1677 		type[0] = 0;
1678 		prom_getprop(node, "device_type", type, sizeof(type));
1679 
1680 		if (type[0] == 0) {
1681 			/*
1682 			 * CHRP Longtrail machines have no device_type
1683 			 * on the memory node, so check the name instead...
1684 			 */
1685 			prom_getprop(node, "name", type, sizeof(type));
1686 		}
1687 		if (prom_strcmp(type, "memory"))
1688 			continue;
1689 
1690 		plen = prom_getprop(node, "reg", regbuf, sizeof(regbuf));
1691 		if (plen > sizeof(regbuf)) {
1692 			prom_printf("memory node too large for buffer !\n");
1693 			plen = sizeof(regbuf);
1694 		}
1695 		p = regbuf;
1696 		endp = p + (plen / sizeof(cell_t));
1697 
1698 #ifdef DEBUG_PROM
1699 		memset(prom_scratch, 0, sizeof(prom_scratch));
1700 		call_prom("package-to-path", 3, 1, node, prom_scratch,
1701 			  sizeof(prom_scratch) - 1);
1702 		prom_debug("  node %s :\n", prom_scratch);
1703 #endif /* DEBUG_PROM */
1704 
1705 		while ((endp - p) >= (rac + rsc)) {
1706 			unsigned long base, size;
1707 
1708 			base = prom_next_cell(rac, &p);
1709 			size = prom_next_cell(rsc, &p);
1710 
1711 			if (size == 0)
1712 				continue;
1713 			prom_debug("    %lx %lx\n", base, size);
1714 			if (base == 0 && (of_platform & PLATFORM_LPAR))
1715 				rmo_top = size;
1716 			if ((base + size) > ram_top)
1717 				ram_top = base + size;
1718 		}
1719 	}
1720 
1721 	alloc_bottom = PAGE_ALIGN((unsigned long)&_end + 0x4000);
1722 
1723 	/*
1724 	 * If prom_memory_limit is set we reduce the upper limits *except* for
1725 	 * alloc_top_high. This must be the real top of RAM so we can put
1726 	 * TCE's up there.
1727 	 */
1728 
1729 	alloc_top_high = ram_top;
1730 
1731 	if (prom_memory_limit) {
1732 		if (prom_memory_limit <= alloc_bottom) {
1733 			prom_printf("Ignoring mem=%lx <= alloc_bottom.\n",
1734 				    prom_memory_limit);
1735 			prom_memory_limit = 0;
1736 		} else if (prom_memory_limit >= ram_top) {
1737 			prom_printf("Ignoring mem=%lx >= ram_top.\n",
1738 				    prom_memory_limit);
1739 			prom_memory_limit = 0;
1740 		} else {
1741 			ram_top = prom_memory_limit;
1742 			rmo_top = min(rmo_top, prom_memory_limit);
1743 		}
1744 	}
1745 
1746 	/*
1747 	 * Setup our top alloc point, that is top of RMO or top of
1748 	 * segment 0 when running non-LPAR.
1749 	 * Some RS64 machines have buggy firmware where claims up at
1750 	 * 1GB fail.  Cap at 768MB as a workaround.
1751 	 * Since 768MB is plenty of room, and we need to cap to something
1752 	 * reasonable on 32-bit, cap at 768MB on all machines.
1753 	 */
1754 	if (!rmo_top)
1755 		rmo_top = ram_top;
1756 	rmo_top = min(0x30000000ul, rmo_top);
1757 	alloc_top = rmo_top;
1758 	alloc_top_high = ram_top;
1759 
1760 	/*
1761 	 * Check if we have an initrd after the kernel but still inside
1762 	 * the RMO.  If we do move our bottom point to after it.
1763 	 */
1764 	if (prom_initrd_start &&
1765 	    prom_initrd_start < rmo_top &&
1766 	    prom_initrd_end > alloc_bottom)
1767 		alloc_bottom = PAGE_ALIGN(prom_initrd_end);
1768 
1769 	prom_printf("memory layout at init:\n");
1770 	prom_printf("  memory_limit : %lx (16 MB aligned)\n",
1771 		    prom_memory_limit);
1772 	prom_printf("  alloc_bottom : %lx\n", alloc_bottom);
1773 	prom_printf("  alloc_top    : %lx\n", alloc_top);
1774 	prom_printf("  alloc_top_hi : %lx\n", alloc_top_high);
1775 	prom_printf("  rmo_top      : %lx\n", rmo_top);
1776 	prom_printf("  ram_top      : %lx\n", ram_top);
1777 }
1778 
1779 static void __init prom_close_stdin(void)
1780 {
1781 	__be32 val;
1782 	ihandle stdin;
1783 
1784 	if (prom_getprop(prom.chosen, "stdin", &val, sizeof(val)) > 0) {
1785 		stdin = be32_to_cpu(val);
1786 		call_prom("close", 1, 0, stdin);
1787 	}
1788 }
1789 
1790 #ifdef CONFIG_PPC_SVM
1791 static int __init prom_rtas_hcall(uint64_t args)
1792 {
1793 	register uint64_t arg1 asm("r3") = H_RTAS;
1794 	register uint64_t arg2 asm("r4") = args;
1795 
1796 	asm volatile("sc 1\n" : "=r" (arg1) :
1797 			"r" (arg1),
1798 			"r" (arg2) :);
1799 	srr_regs_clobbered();
1800 
1801 	return arg1;
1802 }
1803 
1804 static struct rtas_args __prombss os_term_args;
1805 
1806 static void __init prom_rtas_os_term(char *str)
1807 {
1808 	phandle rtas_node;
1809 	__be32 val;
1810 	u32 token;
1811 
1812 	prom_debug("%s: start...\n", __func__);
1813 	rtas_node = call_prom("finddevice", 1, 1, ADDR("/rtas"));
1814 	prom_debug("rtas_node: %x\n", rtas_node);
1815 	if (!PHANDLE_VALID(rtas_node))
1816 		return;
1817 
1818 	val = 0;
1819 	prom_getprop(rtas_node, "ibm,os-term", &val, sizeof(val));
1820 	token = be32_to_cpu(val);
1821 	prom_debug("ibm,os-term: %x\n", token);
1822 	if (token == 0)
1823 		prom_panic("Could not get token for ibm,os-term\n");
1824 	os_term_args.token = cpu_to_be32(token);
1825 	os_term_args.nargs = cpu_to_be32(1);
1826 	os_term_args.nret = cpu_to_be32(1);
1827 	os_term_args.args[0] = cpu_to_be32(__pa(str));
1828 	prom_rtas_hcall((uint64_t)&os_term_args);
1829 }
1830 #endif /* CONFIG_PPC_SVM */
1831 
1832 /*
1833  * Allocate room for and instantiate RTAS
1834  */
1835 static void __init prom_instantiate_rtas(void)
1836 {
1837 	phandle rtas_node;
1838 	ihandle rtas_inst;
1839 	u32 base, entry = 0;
1840 	__be32 val;
1841 	u32 size = 0;
1842 
1843 	prom_debug("prom_instantiate_rtas: start...\n");
1844 
1845 	rtas_node = call_prom("finddevice", 1, 1, ADDR("/rtas"));
1846 	prom_debug("rtas_node: %x\n", rtas_node);
1847 	if (!PHANDLE_VALID(rtas_node))
1848 		return;
1849 
1850 	val = 0;
1851 	prom_getprop(rtas_node, "rtas-size", &val, sizeof(size));
1852 	size = be32_to_cpu(val);
1853 	if (size == 0)
1854 		return;
1855 
1856 	base = alloc_down(size, PAGE_SIZE, 0);
1857 	if (base == 0)
1858 		prom_panic("Could not allocate memory for RTAS\n");
1859 
1860 	rtas_inst = call_prom("open", 1, 1, ADDR("/rtas"));
1861 	if (!IHANDLE_VALID(rtas_inst)) {
1862 		prom_printf("opening rtas package failed (%x)\n", rtas_inst);
1863 		return;
1864 	}
1865 
1866 	prom_printf("instantiating rtas at 0x%x...", base);
1867 
1868 	if (call_prom_ret("call-method", 3, 2, &entry,
1869 			  ADDR("instantiate-rtas"),
1870 			  rtas_inst, base) != 0
1871 	    || entry == 0) {
1872 		prom_printf(" failed\n");
1873 		return;
1874 	}
1875 	prom_printf(" done\n");
1876 
1877 	reserve_mem(base, size);
1878 
1879 	val = cpu_to_be32(base);
1880 	prom_setprop(rtas_node, "/rtas", "linux,rtas-base",
1881 		     &val, sizeof(val));
1882 	val = cpu_to_be32(entry);
1883 	prom_setprop(rtas_node, "/rtas", "linux,rtas-entry",
1884 		     &val, sizeof(val));
1885 
1886 	/* Check if it supports "query-cpu-stopped-state" */
1887 	if (prom_getprop(rtas_node, "query-cpu-stopped-state",
1888 			 &val, sizeof(val)) != PROM_ERROR)
1889 		rtas_has_query_cpu_stopped = true;
1890 
1891 	prom_debug("rtas base     = 0x%x\n", base);
1892 	prom_debug("rtas entry    = 0x%x\n", entry);
1893 	prom_debug("rtas size     = 0x%x\n", size);
1894 
1895 	prom_debug("prom_instantiate_rtas: end...\n");
1896 }
1897 
1898 #ifdef CONFIG_PPC64
1899 /*
1900  * Allocate room for and instantiate Stored Measurement Log (SML)
1901  */
1902 static void __init prom_instantiate_sml(void)
1903 {
1904 	phandle ibmvtpm_node;
1905 	ihandle ibmvtpm_inst;
1906 	u32 entry = 0, size = 0, succ = 0;
1907 	u64 base;
1908 	__be32 val;
1909 
1910 	prom_debug("prom_instantiate_sml: start...\n");
1911 
1912 	ibmvtpm_node = call_prom("finddevice", 1, 1, ADDR("/vdevice/vtpm"));
1913 	prom_debug("ibmvtpm_node: %x\n", ibmvtpm_node);
1914 	if (!PHANDLE_VALID(ibmvtpm_node))
1915 		return;
1916 
1917 	ibmvtpm_inst = call_prom("open", 1, 1, ADDR("/vdevice/vtpm"));
1918 	if (!IHANDLE_VALID(ibmvtpm_inst)) {
1919 		prom_printf("opening vtpm package failed (%x)\n", ibmvtpm_inst);
1920 		return;
1921 	}
1922 
1923 	if (prom_getprop(ibmvtpm_node, "ibm,sml-efi-reformat-supported",
1924 			 &val, sizeof(val)) != PROM_ERROR) {
1925 		if (call_prom_ret("call-method", 2, 2, &succ,
1926 				  ADDR("reformat-sml-to-efi-alignment"),
1927 				  ibmvtpm_inst) != 0 || succ == 0) {
1928 			prom_printf("Reformat SML to EFI alignment failed\n");
1929 			return;
1930 		}
1931 
1932 		if (call_prom_ret("call-method", 2, 2, &size,
1933 				  ADDR("sml-get-allocated-size"),
1934 				  ibmvtpm_inst) != 0 || size == 0) {
1935 			prom_printf("SML get allocated size failed\n");
1936 			return;
1937 		}
1938 	} else {
1939 		if (call_prom_ret("call-method", 2, 2, &size,
1940 				  ADDR("sml-get-handover-size"),
1941 				  ibmvtpm_inst) != 0 || size == 0) {
1942 			prom_printf("SML get handover size failed\n");
1943 			return;
1944 		}
1945 	}
1946 
1947 	base = alloc_down(size, PAGE_SIZE, 0);
1948 	if (base == 0)
1949 		prom_panic("Could not allocate memory for sml\n");
1950 
1951 	prom_printf("instantiating sml at 0x%llx...", base);
1952 
1953 	memset((void *)base, 0, size);
1954 
1955 	if (call_prom_ret("call-method", 4, 2, &entry,
1956 			  ADDR("sml-handover"),
1957 			  ibmvtpm_inst, size, base) != 0 || entry == 0) {
1958 		prom_printf("SML handover failed\n");
1959 		return;
1960 	}
1961 	prom_printf(" done\n");
1962 
1963 	reserve_mem(base, size);
1964 
1965 	prom_setprop(ibmvtpm_node, "/vdevice/vtpm", "linux,sml-base",
1966 		     &base, sizeof(base));
1967 	prom_setprop(ibmvtpm_node, "/vdevice/vtpm", "linux,sml-size",
1968 		     &size, sizeof(size));
1969 
1970 	prom_debug("sml base     = 0x%llx\n", base);
1971 	prom_debug("sml size     = 0x%x\n", size);
1972 
1973 	prom_debug("prom_instantiate_sml: end...\n");
1974 }
1975 
1976 /*
1977  * Allocate room for and initialize TCE tables
1978  */
1979 #ifdef __BIG_ENDIAN__
1980 static void __init prom_initialize_tce_table(void)
1981 {
1982 	phandle node;
1983 	ihandle phb_node;
1984 	char compatible[64], type[64], model[64];
1985 	char *path = prom_scratch;
1986 	u64 base, align;
1987 	u32 minalign, minsize;
1988 	u64 tce_entry, *tce_entryp;
1989 	u64 local_alloc_top, local_alloc_bottom;
1990 	u64 i;
1991 
1992 	if (prom_iommu_off)
1993 		return;
1994 
1995 	prom_debug("starting prom_initialize_tce_table\n");
1996 
1997 	/* Cache current top of allocs so we reserve a single block */
1998 	local_alloc_top = alloc_top_high;
1999 	local_alloc_bottom = local_alloc_top;
2000 
2001 	/* Search all nodes looking for PHBs. */
2002 	for (node = 0; prom_next_node(&node); ) {
2003 		compatible[0] = 0;
2004 		type[0] = 0;
2005 		model[0] = 0;
2006 		prom_getprop(node, "compatible",
2007 			     compatible, sizeof(compatible));
2008 		prom_getprop(node, "device_type", type, sizeof(type));
2009 		prom_getprop(node, "model", model, sizeof(model));
2010 
2011 		if ((type[0] == 0) || (prom_strstr(type, "pci") == NULL))
2012 			continue;
2013 
2014 		/* Keep the old logic intact to avoid regression. */
2015 		if (compatible[0] != 0) {
2016 			if ((prom_strstr(compatible, "python") == NULL) &&
2017 			    (prom_strstr(compatible, "Speedwagon") == NULL) &&
2018 			    (prom_strstr(compatible, "Winnipeg") == NULL))
2019 				continue;
2020 		} else if (model[0] != 0) {
2021 			if ((prom_strstr(model, "ython") == NULL) &&
2022 			    (prom_strstr(model, "peedwagon") == NULL) &&
2023 			    (prom_strstr(model, "innipeg") == NULL))
2024 				continue;
2025 		}
2026 
2027 		if (prom_getprop(node, "tce-table-minalign", &minalign,
2028 				 sizeof(minalign)) == PROM_ERROR)
2029 			minalign = 0;
2030 		if (prom_getprop(node, "tce-table-minsize", &minsize,
2031 				 sizeof(minsize)) == PROM_ERROR)
2032 			minsize = 4UL << 20;
2033 
2034 		/*
2035 		 * Even though we read what OF wants, we just set the table
2036 		 * size to 4 MB.  This is enough to map 2GB of PCI DMA space.
2037 		 * By doing this, we avoid the pitfalls of trying to DMA to
2038 		 * MMIO space and the DMA alias hole.
2039 		 */
2040 		minsize = 4UL << 20;
2041 
2042 		/* Align to the greater of the align or size */
2043 		align = max(minalign, minsize);
2044 		base = alloc_down(minsize, align, 1);
2045 		if (base == 0)
2046 			prom_panic("ERROR, cannot find space for TCE table.\n");
2047 		if (base < local_alloc_bottom)
2048 			local_alloc_bottom = base;
2049 
2050 		/* It seems OF doesn't null-terminate the path :-( */
2051 		memset(path, 0, sizeof(prom_scratch));
2052 		/* Call OF to setup the TCE hardware */
2053 		if (call_prom("package-to-path", 3, 1, node,
2054 			      path, sizeof(prom_scratch) - 1) == PROM_ERROR) {
2055 			prom_printf("package-to-path failed\n");
2056 		}
2057 
2058 		/* Save away the TCE table attributes for later use. */
2059 		prom_setprop(node, path, "linux,tce-base", &base, sizeof(base));
2060 		prom_setprop(node, path, "linux,tce-size", &minsize, sizeof(minsize));
2061 
2062 		prom_debug("TCE table: %s\n", path);
2063 		prom_debug("\tnode = 0x%x\n", node);
2064 		prom_debug("\tbase = 0x%llx\n", base);
2065 		prom_debug("\tsize = 0x%x\n", minsize);
2066 
2067 		/* Initialize the table to have a one-to-one mapping
2068 		 * over the allocated size.
2069 		 */
2070 		tce_entryp = (u64 *)base;
2071 		for (i = 0; i < (minsize >> 3) ;tce_entryp++, i++) {
2072 			tce_entry = (i << PAGE_SHIFT);
2073 			tce_entry |= 0x3;
2074 			*tce_entryp = tce_entry;
2075 		}
2076 
2077 		prom_printf("opening PHB %s", path);
2078 		phb_node = call_prom("open", 1, 1, path);
2079 		if (phb_node == 0)
2080 			prom_printf("... failed\n");
2081 		else
2082 			prom_printf("... done\n");
2083 
2084 		call_prom("call-method", 6, 0, ADDR("set-64-bit-addressing"),
2085 			  phb_node, -1, minsize,
2086 			  (u32) base, (u32) (base >> 32));
2087 		call_prom("close", 1, 0, phb_node);
2088 	}
2089 
2090 	reserve_mem(local_alloc_bottom, local_alloc_top - local_alloc_bottom);
2091 
2092 	/* These are only really needed if there is a memory limit in
2093 	 * effect, but we don't know so export them always. */
2094 	prom_tce_alloc_start = local_alloc_bottom;
2095 	prom_tce_alloc_end = local_alloc_top;
2096 
2097 	/* Flag the first invalid entry */
2098 	prom_debug("ending prom_initialize_tce_table\n");
2099 }
2100 #endif /* __BIG_ENDIAN__ */
2101 #endif /* CONFIG_PPC64 */
2102 
2103 /*
2104  * With CHRP SMP we need to use the OF to start the other processors.
2105  * We can't wait until smp_boot_cpus (the OF is trashed by then)
2106  * so we have to put the processors into a holding pattern controlled
2107  * by the kernel (not OF) before we destroy the OF.
2108  *
2109  * This uses a chunk of low memory, puts some holding pattern
2110  * code there and sends the other processors off to there until
2111  * smp_boot_cpus tells them to do something.  The holding pattern
2112  * checks that address until its cpu # is there, when it is that
2113  * cpu jumps to __secondary_start().  smp_boot_cpus() takes care
2114  * of setting those values.
2115  *
2116  * We also use physical address 0x4 here to tell when a cpu
2117  * is in its holding pattern code.
2118  *
2119  * -- Cort
2120  */
2121 /*
2122  * We want to reference the copy of __secondary_hold_* in the
2123  * 0 - 0x100 address range
2124  */
2125 #define LOW_ADDR(x)	(((unsigned long) &(x)) & 0xff)
2126 
2127 static void __init prom_hold_cpus(void)
2128 {
2129 	unsigned long i;
2130 	phandle node;
2131 	char type[64];
2132 	unsigned long *spinloop
2133 		= (void *) LOW_ADDR(__secondary_hold_spinloop);
2134 	unsigned long *acknowledge
2135 		= (void *) LOW_ADDR(__secondary_hold_acknowledge);
2136 	unsigned long secondary_hold = LOW_ADDR(__secondary_hold);
2137 
2138 	/*
2139 	 * On pseries, if RTAS supports "query-cpu-stopped-state",
2140 	 * we skip this stage, the CPUs will be started by the
2141 	 * kernel using RTAS.
2142 	 */
2143 	if ((of_platform == PLATFORM_PSERIES ||
2144 	     of_platform == PLATFORM_PSERIES_LPAR) &&
2145 	    rtas_has_query_cpu_stopped) {
2146 		prom_printf("prom_hold_cpus: skipped\n");
2147 		return;
2148 	}
2149 
2150 	prom_debug("prom_hold_cpus: start...\n");
2151 	prom_debug("    1) spinloop       = 0x%lx\n", (unsigned long)spinloop);
2152 	prom_debug("    1) *spinloop      = 0x%lx\n", *spinloop);
2153 	prom_debug("    1) acknowledge    = 0x%lx\n",
2154 		   (unsigned long)acknowledge);
2155 	prom_debug("    1) *acknowledge   = 0x%lx\n", *acknowledge);
2156 	prom_debug("    1) secondary_hold = 0x%lx\n", secondary_hold);
2157 
2158 	/* Set the common spinloop variable, so all of the secondary cpus
2159 	 * will block when they are awakened from their OF spinloop.
2160 	 * This must occur for both SMP and non SMP kernels, since OF will
2161 	 * be trashed when we move the kernel.
2162 	 */
2163 	*spinloop = 0;
2164 
2165 	/* look for cpus */
2166 	for (node = 0; prom_next_node(&node); ) {
2167 		unsigned int cpu_no;
2168 		__be32 reg;
2169 
2170 		type[0] = 0;
2171 		prom_getprop(node, "device_type", type, sizeof(type));
2172 		if (prom_strcmp(type, "cpu") != 0)
2173 			continue;
2174 
2175 		/* Skip non-configured cpus. */
2176 		if (prom_getprop(node, "status", type, sizeof(type)) > 0)
2177 			if (prom_strcmp(type, "okay") != 0)
2178 				continue;
2179 
2180 		reg = cpu_to_be32(-1); /* make sparse happy */
2181 		prom_getprop(node, "reg", &reg, sizeof(reg));
2182 		cpu_no = be32_to_cpu(reg);
2183 
2184 		prom_debug("cpu hw idx   = %u\n", cpu_no);
2185 
2186 		/* Init the acknowledge var which will be reset by
2187 		 * the secondary cpu when it awakens from its OF
2188 		 * spinloop.
2189 		 */
2190 		*acknowledge = (unsigned long)-1;
2191 
2192 		if (cpu_no != prom.cpu) {
2193 			/* Primary Thread of non-boot cpu or any thread */
2194 			prom_printf("starting cpu hw idx %u... ", cpu_no);
2195 			call_prom("start-cpu", 3, 0, node,
2196 				  secondary_hold, cpu_no);
2197 
2198 			for (i = 0; (i < 100000000) &&
2199 			     (*acknowledge == ((unsigned long)-1)); i++ )
2200 				mb();
2201 
2202 			if (*acknowledge == cpu_no)
2203 				prom_printf("done\n");
2204 			else
2205 				prom_printf("failed: %lx\n", *acknowledge);
2206 		}
2207 #ifdef CONFIG_SMP
2208 		else
2209 			prom_printf("boot cpu hw idx %u\n", cpu_no);
2210 #endif /* CONFIG_SMP */
2211 	}
2212 
2213 	prom_debug("prom_hold_cpus: end...\n");
2214 }
2215 
2216 
2217 static void __init prom_init_client_services(unsigned long pp)
2218 {
2219 	/* Get a handle to the prom entry point before anything else */
2220 	prom_entry = pp;
2221 
2222 	/* get a handle for the stdout device */
2223 	prom.chosen = call_prom("finddevice", 1, 1, ADDR("/chosen"));
2224 	if (!PHANDLE_VALID(prom.chosen))
2225 		prom_panic("cannot find chosen"); /* msg won't be printed :( */
2226 
2227 	/* get device tree root */
2228 	prom.root = call_prom("finddevice", 1, 1, ADDR("/"));
2229 	if (!PHANDLE_VALID(prom.root))
2230 		prom_panic("cannot find device tree root"); /* msg won't be printed :( */
2231 
2232 	prom.mmumap = 0;
2233 }
2234 
2235 #ifdef CONFIG_PPC32
2236 /*
2237  * For really old powermacs, we need to map things we claim.
2238  * For that, we need the ihandle of the mmu.
2239  * Also, on the longtrail, we need to work around other bugs.
2240  */
2241 static void __init prom_find_mmu(void)
2242 {
2243 	phandle oprom;
2244 	char version[64];
2245 
2246 	oprom = call_prom("finddevice", 1, 1, ADDR("/openprom"));
2247 	if (!PHANDLE_VALID(oprom))
2248 		return;
2249 	if (prom_getprop(oprom, "model", version, sizeof(version)) <= 0)
2250 		return;
2251 	version[sizeof(version) - 1] = 0;
2252 	/* XXX might need to add other versions here */
2253 	if (prom_strcmp(version, "Open Firmware, 1.0.5") == 0)
2254 		of_workarounds = OF_WA_CLAIM;
2255 	else if (prom_strncmp(version, "FirmWorks,3.", 12) == 0) {
2256 		of_workarounds = OF_WA_CLAIM | OF_WA_LONGTRAIL;
2257 		call_prom("interpret", 1, 1, "dev /memory 0 to allow-reclaim");
2258 	} else
2259 		return;
2260 	prom.memory = call_prom("open", 1, 1, ADDR("/memory"));
2261 	prom_getprop(prom.chosen, "mmu", &prom.mmumap,
2262 		     sizeof(prom.mmumap));
2263 	prom.mmumap = be32_to_cpu(prom.mmumap);
2264 	if (!IHANDLE_VALID(prom.memory) || !IHANDLE_VALID(prom.mmumap))
2265 		of_workarounds &= ~OF_WA_CLAIM;		/* hmmm */
2266 }
2267 #else
2268 #define prom_find_mmu()
2269 #endif
2270 
2271 static void __init prom_init_stdout(void)
2272 {
2273 	char *path = of_stdout_device;
2274 	char type[16];
2275 	phandle stdout_node;
2276 	__be32 val;
2277 
2278 	if (prom_getprop(prom.chosen, "stdout", &val, sizeof(val)) <= 0)
2279 		prom_panic("cannot find stdout");
2280 
2281 	prom.stdout = be32_to_cpu(val);
2282 
2283 	/* Get the full OF pathname of the stdout device */
2284 	memset(path, 0, 256);
2285 	call_prom("instance-to-path", 3, 1, prom.stdout, path, 255);
2286 	prom_printf("OF stdout device is: %s\n", of_stdout_device);
2287 	prom_setprop(prom.chosen, "/chosen", "linux,stdout-path",
2288 		     path, prom_strlen(path) + 1);
2289 
2290 	/* instance-to-package fails on PA-Semi */
2291 	stdout_node = call_prom("instance-to-package", 1, 1, prom.stdout);
2292 	if (stdout_node != PROM_ERROR) {
2293 		val = cpu_to_be32(stdout_node);
2294 
2295 		/* If it's a display, note it */
2296 		memset(type, 0, sizeof(type));
2297 		prom_getprop(stdout_node, "device_type", type, sizeof(type));
2298 		if (prom_strcmp(type, "display") == 0)
2299 			prom_setprop(stdout_node, path, "linux,boot-display", NULL, 0);
2300 	}
2301 }
2302 
2303 static int __init prom_find_machine_type(void)
2304 {
2305 	char compat[256];
2306 	int len, i = 0;
2307 #ifdef CONFIG_PPC64
2308 	phandle rtas;
2309 	int x;
2310 #endif
2311 
2312 	/* Look for a PowerMac or a Cell */
2313 	len = prom_getprop(prom.root, "compatible",
2314 			   compat, sizeof(compat)-1);
2315 	if (len > 0) {
2316 		compat[len] = 0;
2317 		while (i < len) {
2318 			char *p = &compat[i];
2319 			int sl = prom_strlen(p);
2320 			if (sl == 0)
2321 				break;
2322 			if (prom_strstr(p, "Power Macintosh") ||
2323 			    prom_strstr(p, "MacRISC"))
2324 				return PLATFORM_POWERMAC;
2325 #ifdef CONFIG_PPC64
2326 			/* We must make sure we don't detect the IBM Cell
2327 			 * blades as pSeries due to some firmware issues,
2328 			 * so we do it here.
2329 			 */
2330 			if (prom_strstr(p, "IBM,CBEA") ||
2331 			    prom_strstr(p, "IBM,CPBW-1.0"))
2332 				return PLATFORM_GENERIC;
2333 #endif /* CONFIG_PPC64 */
2334 			i += sl + 1;
2335 		}
2336 	}
2337 #ifdef CONFIG_PPC64
2338 	/* Try to figure out if it's an IBM pSeries or any other
2339 	 * PAPR compliant platform. We assume it is if :
2340 	 *  - /device_type is "chrp" (please, do NOT use that for future
2341 	 *    non-IBM designs !
2342 	 *  - it has /rtas
2343 	 */
2344 	len = prom_getprop(prom.root, "device_type",
2345 			   compat, sizeof(compat)-1);
2346 	if (len <= 0)
2347 		return PLATFORM_GENERIC;
2348 	if (prom_strcmp(compat, "chrp"))
2349 		return PLATFORM_GENERIC;
2350 
2351 	/* Default to pSeries. We need to know if we are running LPAR */
2352 	rtas = call_prom("finddevice", 1, 1, ADDR("/rtas"));
2353 	if (!PHANDLE_VALID(rtas))
2354 		return PLATFORM_GENERIC;
2355 	x = prom_getproplen(rtas, "ibm,hypertas-functions");
2356 	if (x != PROM_ERROR) {
2357 		prom_debug("Hypertas detected, assuming LPAR !\n");
2358 		return PLATFORM_PSERIES_LPAR;
2359 	}
2360 	return PLATFORM_PSERIES;
2361 #else
2362 	return PLATFORM_GENERIC;
2363 #endif
2364 }
2365 
2366 static int __init prom_set_color(ihandle ih, int i, int r, int g, int b)
2367 {
2368 	return call_prom("call-method", 6, 1, ADDR("color!"), ih, i, b, g, r);
2369 }
2370 
2371 /*
2372  * If we have a display that we don't know how to drive,
2373  * we will want to try to execute OF's open method for it
2374  * later.  However, OF will probably fall over if we do that
2375  * we've taken over the MMU.
2376  * So we check whether we will need to open the display,
2377  * and if so, open it now.
2378  */
2379 static void __init prom_check_displays(void)
2380 {
2381 	char type[16], *path;
2382 	phandle node;
2383 	ihandle ih;
2384 	int i;
2385 
2386 	static const unsigned char default_colors[] __initconst = {
2387 		0x00, 0x00, 0x00,
2388 		0x00, 0x00, 0xaa,
2389 		0x00, 0xaa, 0x00,
2390 		0x00, 0xaa, 0xaa,
2391 		0xaa, 0x00, 0x00,
2392 		0xaa, 0x00, 0xaa,
2393 		0xaa, 0xaa, 0x00,
2394 		0xaa, 0xaa, 0xaa,
2395 		0x55, 0x55, 0x55,
2396 		0x55, 0x55, 0xff,
2397 		0x55, 0xff, 0x55,
2398 		0x55, 0xff, 0xff,
2399 		0xff, 0x55, 0x55,
2400 		0xff, 0x55, 0xff,
2401 		0xff, 0xff, 0x55,
2402 		0xff, 0xff, 0xff
2403 	};
2404 	const unsigned char *clut;
2405 
2406 	prom_debug("Looking for displays\n");
2407 	for (node = 0; prom_next_node(&node); ) {
2408 		memset(type, 0, sizeof(type));
2409 		prom_getprop(node, "device_type", type, sizeof(type));
2410 		if (prom_strcmp(type, "display") != 0)
2411 			continue;
2412 
2413 		/* It seems OF doesn't null-terminate the path :-( */
2414 		path = prom_scratch;
2415 		memset(path, 0, sizeof(prom_scratch));
2416 
2417 		/*
2418 		 * leave some room at the end of the path for appending extra
2419 		 * arguments
2420 		 */
2421 		if (call_prom("package-to-path", 3, 1, node, path,
2422 			      sizeof(prom_scratch) - 10) == PROM_ERROR)
2423 			continue;
2424 		prom_printf("found display   : %s, opening... ", path);
2425 
2426 		ih = call_prom("open", 1, 1, path);
2427 		if (ih == 0) {
2428 			prom_printf("failed\n");
2429 			continue;
2430 		}
2431 
2432 		/* Success */
2433 		prom_printf("done\n");
2434 		prom_setprop(node, path, "linux,opened", NULL, 0);
2435 
2436 		/* Setup a usable color table when the appropriate
2437 		 * method is available. Should update this to set-colors */
2438 		clut = default_colors;
2439 		for (i = 0; i < 16; i++, clut += 3)
2440 			if (prom_set_color(ih, i, clut[0], clut[1],
2441 					   clut[2]) != 0)
2442 				break;
2443 
2444 #ifdef CONFIG_LOGO_LINUX_CLUT224
2445 		clut = PTRRELOC(logo_linux_clut224.clut);
2446 		for (i = 0; i < logo_linux_clut224.clutsize; i++, clut += 3)
2447 			if (prom_set_color(ih, i + 32, clut[0], clut[1],
2448 					   clut[2]) != 0)
2449 				break;
2450 #endif /* CONFIG_LOGO_LINUX_CLUT224 */
2451 
2452 #ifdef CONFIG_PPC_EARLY_DEBUG_BOOTX
2453 		if (prom_getprop(node, "linux,boot-display", NULL, 0) !=
2454 		    PROM_ERROR) {
2455 			u32 width, height, pitch, addr;
2456 
2457 			prom_printf("Setting btext !\n");
2458 
2459 			if (prom_getprop(node, "width", &width, 4) == PROM_ERROR)
2460 				return;
2461 
2462 			if (prom_getprop(node, "height", &height, 4) == PROM_ERROR)
2463 				return;
2464 
2465 			if (prom_getprop(node, "linebytes", &pitch, 4) == PROM_ERROR)
2466 				return;
2467 
2468 			if (prom_getprop(node, "address", &addr, 4) == PROM_ERROR)
2469 				return;
2470 
2471 			prom_printf("W=%d H=%d LB=%d addr=0x%x\n",
2472 				    width, height, pitch, addr);
2473 			btext_setup_display(width, height, 8, pitch, addr);
2474 			btext_prepare_BAT();
2475 		}
2476 #endif /* CONFIG_PPC_EARLY_DEBUG_BOOTX */
2477 	}
2478 }
2479 
2480 
2481 /* Return (relocated) pointer to this much memory: moves initrd if reqd. */
2482 static void __init *make_room(unsigned long *mem_start, unsigned long *mem_end,
2483 			      unsigned long needed, unsigned long align)
2484 {
2485 	void *ret;
2486 
2487 	*mem_start = ALIGN(*mem_start, align);
2488 	while ((*mem_start + needed) > *mem_end) {
2489 		unsigned long room, chunk;
2490 
2491 		prom_debug("Chunk exhausted, claiming more at %lx...\n",
2492 			   alloc_bottom);
2493 		room = alloc_top - alloc_bottom;
2494 		if (room > DEVTREE_CHUNK_SIZE)
2495 			room = DEVTREE_CHUNK_SIZE;
2496 		if (room < PAGE_SIZE)
2497 			prom_panic("No memory for flatten_device_tree "
2498 				   "(no room)\n");
2499 		chunk = alloc_up(room, 0);
2500 		if (chunk == 0)
2501 			prom_panic("No memory for flatten_device_tree "
2502 				   "(claim failed)\n");
2503 		*mem_end = chunk + room;
2504 	}
2505 
2506 	ret = (void *)*mem_start;
2507 	*mem_start += needed;
2508 
2509 	return ret;
2510 }
2511 
2512 #define dt_push_token(token, mem_start, mem_end) do { 			\
2513 		void *room = make_room(mem_start, mem_end, 4, 4);	\
2514 		*(__be32 *)room = cpu_to_be32(token);			\
2515 	} while(0)
2516 
2517 static unsigned long __init dt_find_string(char *str)
2518 {
2519 	char *s, *os;
2520 
2521 	s = os = (char *)dt_string_start;
2522 	s += 4;
2523 	while (s <  (char *)dt_string_end) {
2524 		if (prom_strcmp(s, str) == 0)
2525 			return s - os;
2526 		s += prom_strlen(s) + 1;
2527 	}
2528 	return 0;
2529 }
2530 
2531 /*
2532  * The Open Firmware 1275 specification states properties must be 31 bytes or
2533  * less, however not all firmwares obey this. Make it 64 bytes to be safe.
2534  */
2535 #define MAX_PROPERTY_NAME 64
2536 
2537 static void __init scan_dt_build_strings(phandle node,
2538 					 unsigned long *mem_start,
2539 					 unsigned long *mem_end)
2540 {
2541 	char *prev_name, *namep, *sstart;
2542 	unsigned long soff;
2543 	phandle child;
2544 
2545 	sstart =  (char *)dt_string_start;
2546 
2547 	/* get and store all property names */
2548 	prev_name = "";
2549 	for (;;) {
2550 		/* 64 is max len of name including nul. */
2551 		namep = make_room(mem_start, mem_end, MAX_PROPERTY_NAME, 1);
2552 		if (call_prom("nextprop", 3, 1, node, prev_name, namep) != 1) {
2553 			/* No more nodes: unwind alloc */
2554 			*mem_start = (unsigned long)namep;
2555 			break;
2556 		}
2557 
2558  		/* skip "name" */
2559 		if (prom_strcmp(namep, "name") == 0) {
2560  			*mem_start = (unsigned long)namep;
2561  			prev_name = "name";
2562  			continue;
2563  		}
2564 		/* get/create string entry */
2565 		soff = dt_find_string(namep);
2566 		if (soff != 0) {
2567 			*mem_start = (unsigned long)namep;
2568 			namep = sstart + soff;
2569 		} else {
2570 			/* Trim off some if we can */
2571 			*mem_start = (unsigned long)namep + prom_strlen(namep) + 1;
2572 			dt_string_end = *mem_start;
2573 		}
2574 		prev_name = namep;
2575 	}
2576 
2577 	/* do all our children */
2578 	child = call_prom("child", 1, 1, node);
2579 	while (child != 0) {
2580 		scan_dt_build_strings(child, mem_start, mem_end);
2581 		child = call_prom("peer", 1, 1, child);
2582 	}
2583 }
2584 
2585 static void __init scan_dt_build_struct(phandle node, unsigned long *mem_start,
2586 					unsigned long *mem_end)
2587 {
2588 	phandle child;
2589 	char *namep, *prev_name, *sstart, *p, *ep, *lp, *path;
2590 	unsigned long soff;
2591 	unsigned char *valp;
2592 	static char pname[MAX_PROPERTY_NAME] __prombss;
2593 	int l, room, has_phandle = 0;
2594 
2595 	dt_push_token(OF_DT_BEGIN_NODE, mem_start, mem_end);
2596 
2597 	/* get the node's full name */
2598 	namep = (char *)*mem_start;
2599 	room = *mem_end - *mem_start;
2600 	if (room > 255)
2601 		room = 255;
2602 	l = call_prom("package-to-path", 3, 1, node, namep, room);
2603 	if (l >= 0) {
2604 		/* Didn't fit?  Get more room. */
2605 		if (l >= room) {
2606 			if (l >= *mem_end - *mem_start)
2607 				namep = make_room(mem_start, mem_end, l+1, 1);
2608 			call_prom("package-to-path", 3, 1, node, namep, l);
2609 		}
2610 		namep[l] = '\0';
2611 
2612 		/* Fixup an Apple bug where they have bogus \0 chars in the
2613 		 * middle of the path in some properties, and extract
2614 		 * the unit name (everything after the last '/').
2615 		 */
2616 		for (lp = p = namep, ep = namep + l; p < ep; p++) {
2617 			if (*p == '/')
2618 				lp = namep;
2619 			else if (*p != 0)
2620 				*lp++ = *p;
2621 		}
2622 		*lp = 0;
2623 		*mem_start = ALIGN((unsigned long)lp + 1, 4);
2624 	}
2625 
2626 	/* get it again for debugging */
2627 	path = prom_scratch;
2628 	memset(path, 0, sizeof(prom_scratch));
2629 	call_prom("package-to-path", 3, 1, node, path, sizeof(prom_scratch) - 1);
2630 
2631 	/* get and store all properties */
2632 	prev_name = "";
2633 	sstart = (char *)dt_string_start;
2634 	for (;;) {
2635 		if (call_prom("nextprop", 3, 1, node, prev_name,
2636 			      pname) != 1)
2637 			break;
2638 
2639  		/* skip "name" */
2640 		if (prom_strcmp(pname, "name") == 0) {
2641  			prev_name = "name";
2642  			continue;
2643  		}
2644 
2645 		/* find string offset */
2646 		soff = dt_find_string(pname);
2647 		if (soff == 0) {
2648 			prom_printf("WARNING: Can't find string index for"
2649 				    " <%s>, node %s\n", pname, path);
2650 			break;
2651 		}
2652 		prev_name = sstart + soff;
2653 
2654 		/* get length */
2655 		l = call_prom("getproplen", 2, 1, node, pname);
2656 
2657 		/* sanity checks */
2658 		if (l == PROM_ERROR)
2659 			continue;
2660 
2661 		/* push property head */
2662 		dt_push_token(OF_DT_PROP, mem_start, mem_end);
2663 		dt_push_token(l, mem_start, mem_end);
2664 		dt_push_token(soff, mem_start, mem_end);
2665 
2666 		/* push property content */
2667 		valp = make_room(mem_start, mem_end, l, 4);
2668 		call_prom("getprop", 4, 1, node, pname, valp, l);
2669 		*mem_start = ALIGN(*mem_start, 4);
2670 
2671 		if (!prom_strcmp(pname, "phandle"))
2672 			has_phandle = 1;
2673 	}
2674 
2675 	/* Add a "phandle" property if none already exist */
2676 	if (!has_phandle) {
2677 		soff = dt_find_string("phandle");
2678 		if (soff == 0)
2679 			prom_printf("WARNING: Can't find string index for <phandle> node %s\n", path);
2680 		else {
2681 			dt_push_token(OF_DT_PROP, mem_start, mem_end);
2682 			dt_push_token(4, mem_start, mem_end);
2683 			dt_push_token(soff, mem_start, mem_end);
2684 			valp = make_room(mem_start, mem_end, 4, 4);
2685 			*(__be32 *)valp = cpu_to_be32(node);
2686 		}
2687 	}
2688 
2689 	/* do all our children */
2690 	child = call_prom("child", 1, 1, node);
2691 	while (child != 0) {
2692 		scan_dt_build_struct(child, mem_start, mem_end);
2693 		child = call_prom("peer", 1, 1, child);
2694 	}
2695 
2696 	dt_push_token(OF_DT_END_NODE, mem_start, mem_end);
2697 }
2698 
2699 static void __init flatten_device_tree(void)
2700 {
2701 	phandle root;
2702 	unsigned long mem_start, mem_end, room;
2703 	struct boot_param_header *hdr;
2704 	char *namep;
2705 	u64 *rsvmap;
2706 
2707 	/*
2708 	 * Check how much room we have between alloc top & bottom (+/- a
2709 	 * few pages), crop to 1MB, as this is our "chunk" size
2710 	 */
2711 	room = alloc_top - alloc_bottom - 0x4000;
2712 	if (room > DEVTREE_CHUNK_SIZE)
2713 		room = DEVTREE_CHUNK_SIZE;
2714 	prom_debug("starting device tree allocs at %lx\n", alloc_bottom);
2715 
2716 	/* Now try to claim that */
2717 	mem_start = (unsigned long)alloc_up(room, PAGE_SIZE);
2718 	if (mem_start == 0)
2719 		prom_panic("Can't allocate initial device-tree chunk\n");
2720 	mem_end = mem_start + room;
2721 
2722 	/* Get root of tree */
2723 	root = call_prom("peer", 1, 1, (phandle)0);
2724 	if (root == (phandle)0)
2725 		prom_panic ("couldn't get device tree root\n");
2726 
2727 	/* Build header and make room for mem rsv map */
2728 	mem_start = ALIGN(mem_start, 4);
2729 	hdr = make_room(&mem_start, &mem_end,
2730 			sizeof(struct boot_param_header), 4);
2731 	dt_header_start = (unsigned long)hdr;
2732 	rsvmap = make_room(&mem_start, &mem_end, sizeof(mem_reserve_map), 8);
2733 
2734 	/* Start of strings */
2735 	mem_start = PAGE_ALIGN(mem_start);
2736 	dt_string_start = mem_start;
2737 	mem_start += 4; /* hole */
2738 
2739 	/* Add "phandle" in there, we'll need it */
2740 	namep = make_room(&mem_start, &mem_end, 16, 1);
2741 	prom_strscpy_pad(namep, "phandle", sizeof("phandle"));
2742 	mem_start = (unsigned long)namep + prom_strlen(namep) + 1;
2743 
2744 	/* Build string array */
2745 	prom_printf("Building dt strings...\n");
2746 	scan_dt_build_strings(root, &mem_start, &mem_end);
2747 	dt_string_end = mem_start;
2748 
2749 	/* Build structure */
2750 	mem_start = PAGE_ALIGN(mem_start);
2751 	dt_struct_start = mem_start;
2752 	prom_printf("Building dt structure...\n");
2753 	scan_dt_build_struct(root, &mem_start, &mem_end);
2754 	dt_push_token(OF_DT_END, &mem_start, &mem_end);
2755 	dt_struct_end = PAGE_ALIGN(mem_start);
2756 
2757 	/* Finish header */
2758 	hdr->boot_cpuid_phys = cpu_to_be32(prom.cpu);
2759 	hdr->magic = cpu_to_be32(OF_DT_HEADER);
2760 	hdr->totalsize = cpu_to_be32(dt_struct_end - dt_header_start);
2761 	hdr->off_dt_struct = cpu_to_be32(dt_struct_start - dt_header_start);
2762 	hdr->off_dt_strings = cpu_to_be32(dt_string_start - dt_header_start);
2763 	hdr->dt_strings_size = cpu_to_be32(dt_string_end - dt_string_start);
2764 	hdr->off_mem_rsvmap = cpu_to_be32(((unsigned long)rsvmap) - dt_header_start);
2765 	hdr->version = cpu_to_be32(OF_DT_VERSION);
2766 	/* Version 16 is not backward compatible */
2767 	hdr->last_comp_version = cpu_to_be32(0x10);
2768 
2769 	/* Copy the reserve map in */
2770 	memcpy(rsvmap, mem_reserve_map, sizeof(mem_reserve_map));
2771 
2772 #ifdef DEBUG_PROM
2773 	{
2774 		int i;
2775 		prom_printf("reserved memory map:\n");
2776 		for (i = 0; i < mem_reserve_cnt; i++)
2777 			prom_printf("  %llx - %llx\n",
2778 				    be64_to_cpu(mem_reserve_map[i].base),
2779 				    be64_to_cpu(mem_reserve_map[i].size));
2780 	}
2781 #endif
2782 	/* Bump mem_reserve_cnt to cause further reservations to fail
2783 	 * since it's too late.
2784 	 */
2785 	mem_reserve_cnt = MEM_RESERVE_MAP_SIZE;
2786 
2787 	prom_printf("Device tree strings 0x%lx -> 0x%lx\n",
2788 		    dt_string_start, dt_string_end);
2789 	prom_printf("Device tree struct  0x%lx -> 0x%lx\n",
2790 		    dt_struct_start, dt_struct_end);
2791 }
2792 
2793 #ifdef CONFIG_PPC_MAPLE
2794 /* PIBS Version 1.05.0000 04/26/2005 has an incorrect /ht/isa/ranges property.
2795  * The values are bad, and it doesn't even have the right number of cells. */
2796 static void __init fixup_device_tree_maple(void)
2797 {
2798 	phandle isa;
2799 	u32 rloc = 0x01002000; /* IO space; PCI device = 4 */
2800 	u32 isa_ranges[6];
2801 	char *name;
2802 
2803 	name = "/ht@0/isa@4";
2804 	isa = call_prom("finddevice", 1, 1, ADDR(name));
2805 	if (!PHANDLE_VALID(isa)) {
2806 		name = "/ht@0/isa@6";
2807 		isa = call_prom("finddevice", 1, 1, ADDR(name));
2808 		rloc = 0x01003000; /* IO space; PCI device = 6 */
2809 	}
2810 	if (!PHANDLE_VALID(isa))
2811 		return;
2812 
2813 	if (prom_getproplen(isa, "ranges") != 12)
2814 		return;
2815 	if (prom_getprop(isa, "ranges", isa_ranges, sizeof(isa_ranges))
2816 		== PROM_ERROR)
2817 		return;
2818 
2819 	if (isa_ranges[0] != 0x1 ||
2820 		isa_ranges[1] != 0xf4000000 ||
2821 		isa_ranges[2] != 0x00010000)
2822 		return;
2823 
2824 	prom_printf("Fixing up bogus ISA range on Maple/Apache...\n");
2825 
2826 	isa_ranges[0] = 0x1;
2827 	isa_ranges[1] = 0x0;
2828 	isa_ranges[2] = rloc;
2829 	isa_ranges[3] = 0x0;
2830 	isa_ranges[4] = 0x0;
2831 	isa_ranges[5] = 0x00010000;
2832 	prom_setprop(isa, name, "ranges",
2833 			isa_ranges, sizeof(isa_ranges));
2834 }
2835 
2836 #define CPC925_MC_START		0xf8000000
2837 #define CPC925_MC_LENGTH	0x1000000
2838 /* The values for memory-controller don't have right number of cells */
2839 static void __init fixup_device_tree_maple_memory_controller(void)
2840 {
2841 	phandle mc;
2842 	u32 mc_reg[4];
2843 	char *name = "/hostbridge@f8000000";
2844 	u32 ac, sc;
2845 
2846 	mc = call_prom("finddevice", 1, 1, ADDR(name));
2847 	if (!PHANDLE_VALID(mc))
2848 		return;
2849 
2850 	if (prom_getproplen(mc, "reg") != 8)
2851 		return;
2852 
2853 	prom_getprop(prom.root, "#address-cells", &ac, sizeof(ac));
2854 	prom_getprop(prom.root, "#size-cells", &sc, sizeof(sc));
2855 	if ((ac != 2) || (sc != 2))
2856 		return;
2857 
2858 	if (prom_getprop(mc, "reg", mc_reg, sizeof(mc_reg)) == PROM_ERROR)
2859 		return;
2860 
2861 	if (mc_reg[0] != CPC925_MC_START || mc_reg[1] != CPC925_MC_LENGTH)
2862 		return;
2863 
2864 	prom_printf("Fixing up bogus hostbridge on Maple...\n");
2865 
2866 	mc_reg[0] = 0x0;
2867 	mc_reg[1] = CPC925_MC_START;
2868 	mc_reg[2] = 0x0;
2869 	mc_reg[3] = CPC925_MC_LENGTH;
2870 	prom_setprop(mc, name, "reg", mc_reg, sizeof(mc_reg));
2871 }
2872 #else
2873 #define fixup_device_tree_maple()
2874 #define fixup_device_tree_maple_memory_controller()
2875 #endif
2876 
2877 #ifdef CONFIG_PPC_CHRP
2878 /*
2879  * Pegasos and BriQ lacks the "ranges" property in the isa node
2880  * Pegasos needs decimal IRQ 14/15, not hexadecimal
2881  * Pegasos has the IDE configured in legacy mode, but advertised as native
2882  */
2883 static void __init fixup_device_tree_chrp(void)
2884 {
2885 	phandle ph;
2886 	u32 prop[6];
2887 	u32 rloc = 0x01006000; /* IO space; PCI device = 12 */
2888 	char *name;
2889 	int rc;
2890 
2891 	name = "/pci@80000000/isa@c";
2892 	ph = call_prom("finddevice", 1, 1, ADDR(name));
2893 	if (!PHANDLE_VALID(ph)) {
2894 		name = "/pci@ff500000/isa@6";
2895 		ph = call_prom("finddevice", 1, 1, ADDR(name));
2896 		rloc = 0x01003000; /* IO space; PCI device = 6 */
2897 	}
2898 	if (PHANDLE_VALID(ph)) {
2899 		rc = prom_getproplen(ph, "ranges");
2900 		if (rc == 0 || rc == PROM_ERROR) {
2901 			prom_printf("Fixing up missing ISA range on Pegasos...\n");
2902 
2903 			prop[0] = 0x1;
2904 			prop[1] = 0x0;
2905 			prop[2] = rloc;
2906 			prop[3] = 0x0;
2907 			prop[4] = 0x0;
2908 			prop[5] = 0x00010000;
2909 			prom_setprop(ph, name, "ranges", prop, sizeof(prop));
2910 		}
2911 	}
2912 
2913 	name = "/pci@80000000/ide@C,1";
2914 	ph = call_prom("finddevice", 1, 1, ADDR(name));
2915 	if (PHANDLE_VALID(ph)) {
2916 		prom_printf("Fixing up IDE interrupt on Pegasos...\n");
2917 		prop[0] = 14;
2918 		prop[1] = 0x0;
2919 		prom_setprop(ph, name, "interrupts", prop, 2*sizeof(u32));
2920 		prom_printf("Fixing up IDE class-code on Pegasos...\n");
2921 		rc = prom_getprop(ph, "class-code", prop, sizeof(u32));
2922 		if (rc == sizeof(u32)) {
2923 			prop[0] &= ~0x5;
2924 			prom_setprop(ph, name, "class-code", prop, sizeof(u32));
2925 		}
2926 	}
2927 }
2928 #else
2929 #define fixup_device_tree_chrp()
2930 #endif
2931 
2932 #if defined(CONFIG_PPC64) && defined(CONFIG_PPC_PMAC)
2933 static void __init fixup_device_tree_pmac(void)
2934 {
2935 	phandle u3, i2c, mpic;
2936 	u32 u3_rev;
2937 	u32 interrupts[2];
2938 	u32 parent;
2939 
2940 	/* Some G5s have a missing interrupt definition, fix it up here */
2941 	u3 = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000"));
2942 	if (!PHANDLE_VALID(u3))
2943 		return;
2944 	i2c = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/i2c@f8001000"));
2945 	if (!PHANDLE_VALID(i2c))
2946 		return;
2947 	mpic = call_prom("finddevice", 1, 1, ADDR("/u3@0,f8000000/mpic@f8040000"));
2948 	if (!PHANDLE_VALID(mpic))
2949 		return;
2950 
2951 	/* check if proper rev of u3 */
2952 	if (prom_getprop(u3, "device-rev", &u3_rev, sizeof(u3_rev))
2953 	    == PROM_ERROR)
2954 		return;
2955 	if (u3_rev < 0x35 || u3_rev > 0x39)
2956 		return;
2957 	/* does it need fixup ? */
2958 	if (prom_getproplen(i2c, "interrupts") > 0)
2959 		return;
2960 
2961 	prom_printf("fixing up bogus interrupts for u3 i2c...\n");
2962 
2963 	/* interrupt on this revision of u3 is number 0 and level */
2964 	interrupts[0] = 0;
2965 	interrupts[1] = 1;
2966 	prom_setprop(i2c, "/u3@0,f8000000/i2c@f8001000", "interrupts",
2967 		     &interrupts, sizeof(interrupts));
2968 	parent = (u32)mpic;
2969 	prom_setprop(i2c, "/u3@0,f8000000/i2c@f8001000", "interrupt-parent",
2970 		     &parent, sizeof(parent));
2971 }
2972 #else
2973 #define fixup_device_tree_pmac()
2974 #endif
2975 
2976 #ifdef CONFIG_PPC_EFIKA
2977 /*
2978  * The MPC5200 FEC driver requires an phy-handle property to tell it how
2979  * to talk to the phy.  If the phy-handle property is missing, then this
2980  * function is called to add the appropriate nodes and link it to the
2981  * ethernet node.
2982  */
2983 static void __init fixup_device_tree_efika_add_phy(void)
2984 {
2985 	u32 node;
2986 	char prop[64];
2987 	int rv;
2988 
2989 	/* Check if /builtin/ethernet exists - bail if it doesn't */
2990 	node = call_prom("finddevice", 1, 1, ADDR("/builtin/ethernet"));
2991 	if (!PHANDLE_VALID(node))
2992 		return;
2993 
2994 	/* Check if the phy-handle property exists - bail if it does */
2995 	rv = prom_getprop(node, "phy-handle", prop, sizeof(prop));
2996 	if (rv <= 0)
2997 		return;
2998 
2999 	/*
3000 	 * At this point the ethernet device doesn't have a phy described.
3001 	 * Now we need to add the missing phy node and linkage
3002 	 */
3003 
3004 	/* Check for an MDIO bus node - if missing then create one */
3005 	node = call_prom("finddevice", 1, 1, ADDR("/builtin/mdio"));
3006 	if (!PHANDLE_VALID(node)) {
3007 		prom_printf("Adding Ethernet MDIO node\n");
3008 		call_prom("interpret", 1, 1,
3009 			" s\" /builtin\" find-device"
3010 			" new-device"
3011 				" 1 encode-int s\" #address-cells\" property"
3012 				" 0 encode-int s\" #size-cells\" property"
3013 				" s\" mdio\" device-name"
3014 				" s\" fsl,mpc5200b-mdio\" encode-string"
3015 				" s\" compatible\" property"
3016 				" 0xf0003000 0x400 reg"
3017 				" 0x2 encode-int"
3018 				" 0x5 encode-int encode+"
3019 				" 0x3 encode-int encode+"
3020 				" s\" interrupts\" property"
3021 			" finish-device");
3022 	}
3023 
3024 	/* Check for a PHY device node - if missing then create one and
3025 	 * give it's phandle to the ethernet node */
3026 	node = call_prom("finddevice", 1, 1,
3027 			 ADDR("/builtin/mdio/ethernet-phy"));
3028 	if (!PHANDLE_VALID(node)) {
3029 		prom_printf("Adding Ethernet PHY node\n");
3030 		call_prom("interpret", 1, 1,
3031 			" s\" /builtin/mdio\" find-device"
3032 			" new-device"
3033 				" s\" ethernet-phy\" device-name"
3034 				" 0x10 encode-int s\" reg\" property"
3035 				" my-self"
3036 				" ihandle>phandle"
3037 			" finish-device"
3038 			" s\" /builtin/ethernet\" find-device"
3039 				" encode-int"
3040 				" s\" phy-handle\" property"
3041 			" device-end");
3042 	}
3043 }
3044 
3045 static void __init fixup_device_tree_efika(void)
3046 {
3047 	int sound_irq[3] = { 2, 2, 0 };
3048 	int bcomm_irq[3*16] = { 3,0,0, 3,1,0, 3,2,0, 3,3,0,
3049 				3,4,0, 3,5,0, 3,6,0, 3,7,0,
3050 				3,8,0, 3,9,0, 3,10,0, 3,11,0,
3051 				3,12,0, 3,13,0, 3,14,0, 3,15,0 };
3052 	u32 node;
3053 	char prop[64];
3054 	int rv, len;
3055 
3056 	/* Check if we're really running on a EFIKA */
3057 	node = call_prom("finddevice", 1, 1, ADDR("/"));
3058 	if (!PHANDLE_VALID(node))
3059 		return;
3060 
3061 	rv = prom_getprop(node, "model", prop, sizeof(prop));
3062 	if (rv == PROM_ERROR)
3063 		return;
3064 	if (prom_strcmp(prop, "EFIKA5K2"))
3065 		return;
3066 
3067 	prom_printf("Applying EFIKA device tree fixups\n");
3068 
3069 	/* Claiming to be 'chrp' is death */
3070 	node = call_prom("finddevice", 1, 1, ADDR("/"));
3071 	rv = prom_getprop(node, "device_type", prop, sizeof(prop));
3072 	if (rv != PROM_ERROR && (prom_strcmp(prop, "chrp") == 0))
3073 		prom_setprop(node, "/", "device_type", "efika", sizeof("efika"));
3074 
3075 	/* CODEGEN,description is exposed in /proc/cpuinfo so
3076 	   fix that too */
3077 	rv = prom_getprop(node, "CODEGEN,description", prop, sizeof(prop));
3078 	if (rv != PROM_ERROR && (prom_strstr(prop, "CHRP")))
3079 		prom_setprop(node, "/", "CODEGEN,description",
3080 			     "Efika 5200B PowerPC System",
3081 			     sizeof("Efika 5200B PowerPC System"));
3082 
3083 	/* Fixup bestcomm interrupts property */
3084 	node = call_prom("finddevice", 1, 1, ADDR("/builtin/bestcomm"));
3085 	if (PHANDLE_VALID(node)) {
3086 		len = prom_getproplen(node, "interrupts");
3087 		if (len == 12) {
3088 			prom_printf("Fixing bestcomm interrupts property\n");
3089 			prom_setprop(node, "/builtin/bestcom", "interrupts",
3090 				     bcomm_irq, sizeof(bcomm_irq));
3091 		}
3092 	}
3093 
3094 	/* Fixup sound interrupts property */
3095 	node = call_prom("finddevice", 1, 1, ADDR("/builtin/sound"));
3096 	if (PHANDLE_VALID(node)) {
3097 		rv = prom_getprop(node, "interrupts", prop, sizeof(prop));
3098 		if (rv == PROM_ERROR) {
3099 			prom_printf("Adding sound interrupts property\n");
3100 			prom_setprop(node, "/builtin/sound", "interrupts",
3101 				     sound_irq, sizeof(sound_irq));
3102 		}
3103 	}
3104 
3105 	/* Make sure ethernet phy-handle property exists */
3106 	fixup_device_tree_efika_add_phy();
3107 }
3108 #else
3109 #define fixup_device_tree_efika()
3110 #endif
3111 
3112 #ifdef CONFIG_PPC_PASEMI_NEMO
3113 /*
3114  * CFE supplied on Nemo is broken in several ways, biggest
3115  * problem is that it reassigns ISA interrupts to unused mpic ints.
3116  * Add an interrupt-controller property for the io-bridge to use
3117  * and correct the ints so we can attach them to an irq_domain
3118  */
3119 static void __init fixup_device_tree_pasemi(void)
3120 {
3121 	u32 interrupts[2], parent, rval, val = 0;
3122 	char *name, *pci_name;
3123 	phandle iob, node;
3124 
3125 	/* Find the root pci node */
3126 	name = "/pxp@0,e0000000";
3127 	iob = call_prom("finddevice", 1, 1, ADDR(name));
3128 	if (!PHANDLE_VALID(iob))
3129 		return;
3130 
3131 	/* check if interrupt-controller node set yet */
3132 	if (prom_getproplen(iob, "interrupt-controller") !=PROM_ERROR)
3133 		return;
3134 
3135 	prom_printf("adding interrupt-controller property for SB600...\n");
3136 
3137 	prom_setprop(iob, name, "interrupt-controller", &val, 0);
3138 
3139 	pci_name = "/pxp@0,e0000000/pci@11";
3140 	node = call_prom("finddevice", 1, 1, ADDR(pci_name));
3141 	parent = ADDR(iob);
3142 
3143 	for( ; prom_next_node(&node); ) {
3144 		/* scan each node for one with an interrupt */
3145 		if (!PHANDLE_VALID(node))
3146 			continue;
3147 
3148 		rval = prom_getproplen(node, "interrupts");
3149 		if (rval == 0 || rval == PROM_ERROR)
3150 			continue;
3151 
3152 		prom_getprop(node, "interrupts", &interrupts, sizeof(interrupts));
3153 		if ((interrupts[0] < 212) || (interrupts[0] > 222))
3154 			continue;
3155 
3156 		/* found a node, update both interrupts and interrupt-parent */
3157 		if ((interrupts[0] >= 212) && (interrupts[0] <= 215))
3158 			interrupts[0] -= 203;
3159 		if ((interrupts[0] >= 216) && (interrupts[0] <= 220))
3160 			interrupts[0] -= 213;
3161 		if (interrupts[0] == 221)
3162 			interrupts[0] = 14;
3163 		if (interrupts[0] == 222)
3164 			interrupts[0] = 8;
3165 
3166 		prom_setprop(node, pci_name, "interrupts", interrupts,
3167 					sizeof(interrupts));
3168 		prom_setprop(node, pci_name, "interrupt-parent", &parent,
3169 					sizeof(parent));
3170 	}
3171 
3172 	/*
3173 	 * The io-bridge has device_type set to 'io-bridge' change it to 'isa'
3174 	 * so that generic isa-bridge code can add the SB600 and its on-board
3175 	 * peripherals.
3176 	 */
3177 	name = "/pxp@0,e0000000/io-bridge@0";
3178 	iob = call_prom("finddevice", 1, 1, ADDR(name));
3179 	if (!PHANDLE_VALID(iob))
3180 		return;
3181 
3182 	/* device_type is already set, just change it. */
3183 
3184 	prom_printf("Changing device_type of SB600 node...\n");
3185 
3186 	prom_setprop(iob, name, "device_type", "isa", sizeof("isa"));
3187 }
3188 #else	/* !CONFIG_PPC_PASEMI_NEMO */
3189 static inline void fixup_device_tree_pasemi(void) { }
3190 #endif
3191 
3192 static void __init fixup_device_tree(void)
3193 {
3194 	fixup_device_tree_maple();
3195 	fixup_device_tree_maple_memory_controller();
3196 	fixup_device_tree_chrp();
3197 	fixup_device_tree_pmac();
3198 	fixup_device_tree_efika();
3199 	fixup_device_tree_pasemi();
3200 }
3201 
3202 static void __init prom_find_boot_cpu(void)
3203 {
3204 	__be32 rval;
3205 	ihandle prom_cpu;
3206 	phandle cpu_pkg;
3207 
3208 	rval = 0;
3209 	if (prom_getprop(prom.chosen, "cpu", &rval, sizeof(rval)) <= 0)
3210 		return;
3211 	prom_cpu = be32_to_cpu(rval);
3212 
3213 	cpu_pkg = call_prom("instance-to-package", 1, 1, prom_cpu);
3214 
3215 	if (!PHANDLE_VALID(cpu_pkg))
3216 		return;
3217 
3218 	prom_getprop(cpu_pkg, "reg", &rval, sizeof(rval));
3219 	prom.cpu = be32_to_cpu(rval);
3220 
3221 	prom_debug("Booting CPU hw index = %d\n", prom.cpu);
3222 }
3223 
3224 static void __init prom_check_initrd(unsigned long r3, unsigned long r4)
3225 {
3226 #ifdef CONFIG_BLK_DEV_INITRD
3227 	if (r3 && r4 && r4 != 0xdeadbeef) {
3228 		__be64 val;
3229 
3230 		prom_initrd_start = is_kernel_addr(r3) ? __pa(r3) : r3;
3231 		prom_initrd_end = prom_initrd_start + r4;
3232 
3233 		val = cpu_to_be64(prom_initrd_start);
3234 		prom_setprop(prom.chosen, "/chosen", "linux,initrd-start",
3235 			     &val, sizeof(val));
3236 		val = cpu_to_be64(prom_initrd_end);
3237 		prom_setprop(prom.chosen, "/chosen", "linux,initrd-end",
3238 			     &val, sizeof(val));
3239 
3240 		reserve_mem(prom_initrd_start,
3241 			    prom_initrd_end - prom_initrd_start);
3242 
3243 		prom_debug("initrd_start=0x%lx\n", prom_initrd_start);
3244 		prom_debug("initrd_end=0x%lx\n", prom_initrd_end);
3245 	}
3246 #endif /* CONFIG_BLK_DEV_INITRD */
3247 }
3248 
3249 #ifdef CONFIG_PPC_SVM
3250 /*
3251  * Perform the Enter Secure Mode ultracall.
3252  */
3253 static int __init enter_secure_mode(unsigned long kbase, unsigned long fdt)
3254 {
3255 	register unsigned long r3 asm("r3") = UV_ESM;
3256 	register unsigned long r4 asm("r4") = kbase;
3257 	register unsigned long r5 asm("r5") = fdt;
3258 
3259 	asm volatile("sc 2" : "+r"(r3) : "r"(r4), "r"(r5));
3260 
3261 	return r3;
3262 }
3263 
3264 /*
3265  * Call the Ultravisor to transfer us to secure memory if we have an ESM blob.
3266  */
3267 static void __init setup_secure_guest(unsigned long kbase, unsigned long fdt)
3268 {
3269 	int ret;
3270 
3271 	if (!prom_svm_enable)
3272 		return;
3273 
3274 	/* Switch to secure mode. */
3275 	prom_printf("Switching to secure mode.\n");
3276 
3277 	/*
3278 	 * The ultravisor will do an integrity check of the kernel image but we
3279 	 * relocated it so the check will fail. Restore the original image by
3280 	 * relocating it back to the kernel virtual base address.
3281 	 */
3282 	relocate(KERNELBASE);
3283 
3284 	ret = enter_secure_mode(kbase, fdt);
3285 
3286 	/* Relocate the kernel again. */
3287 	relocate(kbase);
3288 
3289 	if (ret != U_SUCCESS) {
3290 		prom_printf("Returned %d from switching to secure mode.\n", ret);
3291 		prom_rtas_os_term("Switch to secure mode failed.\n");
3292 	}
3293 }
3294 #else
3295 static void __init setup_secure_guest(unsigned long kbase, unsigned long fdt)
3296 {
3297 }
3298 #endif /* CONFIG_PPC_SVM */
3299 
3300 /*
3301  * We enter here early on, when the Open Firmware prom is still
3302  * handling exceptions and the MMU hash table for us.
3303  */
3304 
3305 unsigned long __init prom_init(unsigned long r3, unsigned long r4,
3306 			       unsigned long pp,
3307 			       unsigned long r6, unsigned long r7,
3308 			       unsigned long kbase)
3309 {
3310 	unsigned long hdr;
3311 
3312 #ifdef CONFIG_PPC32
3313 	unsigned long offset = reloc_offset();
3314 	reloc_got2(offset);
3315 #endif
3316 
3317 	/*
3318 	 * First zero the BSS
3319 	 */
3320 	memset(&__bss_start, 0, __bss_stop - __bss_start);
3321 
3322 	/*
3323 	 * Init interface to Open Firmware, get some node references,
3324 	 * like /chosen
3325 	 */
3326 	prom_init_client_services(pp);
3327 
3328 	/*
3329 	 * See if this OF is old enough that we need to do explicit maps
3330 	 * and other workarounds
3331 	 */
3332 	prom_find_mmu();
3333 
3334 	/*
3335 	 * Init prom stdout device
3336 	 */
3337 	prom_init_stdout();
3338 
3339 	prom_printf("Preparing to boot %s", linux_banner);
3340 
3341 	/*
3342 	 * Get default machine type. At this point, we do not differentiate
3343 	 * between pSeries SMP and pSeries LPAR
3344 	 */
3345 	of_platform = prom_find_machine_type();
3346 	prom_printf("Detected machine type: %x\n", of_platform);
3347 
3348 #ifndef CONFIG_NONSTATIC_KERNEL
3349 	/* Bail if this is a kdump kernel. */
3350 	if (PHYSICAL_START > 0)
3351 		prom_panic("Error: You can't boot a kdump kernel from OF!\n");
3352 #endif
3353 
3354 	/*
3355 	 * Check for an initrd
3356 	 */
3357 	prom_check_initrd(r3, r4);
3358 
3359 	/*
3360 	 * Do early parsing of command line
3361 	 */
3362 	early_cmdline_parse();
3363 
3364 #ifdef CONFIG_PPC_PSERIES
3365 	/*
3366 	 * On pSeries, inform the firmware about our capabilities
3367 	 */
3368 	if (of_platform == PLATFORM_PSERIES ||
3369 	    of_platform == PLATFORM_PSERIES_LPAR)
3370 		prom_send_capabilities();
3371 #endif
3372 
3373 	/*
3374 	 * Copy the CPU hold code
3375 	 */
3376 	if (of_platform != PLATFORM_POWERMAC)
3377 		copy_and_flush(0, kbase, 0x100, 0);
3378 
3379 	/*
3380 	 * Initialize memory management within prom_init
3381 	 */
3382 	prom_init_mem();
3383 
3384 	/*
3385 	 * Determine which cpu is actually running right _now_
3386 	 */
3387 	prom_find_boot_cpu();
3388 
3389 	/*
3390 	 * Initialize display devices
3391 	 */
3392 	prom_check_displays();
3393 
3394 #if defined(CONFIG_PPC64) && defined(__BIG_ENDIAN__)
3395 	/*
3396 	 * Initialize IOMMU (TCE tables) on pSeries. Do that before anything else
3397 	 * that uses the allocator, we need to make sure we get the top of memory
3398 	 * available for us here...
3399 	 */
3400 	if (of_platform == PLATFORM_PSERIES)
3401 		prom_initialize_tce_table();
3402 #endif
3403 
3404 	/*
3405 	 * On non-powermacs, try to instantiate RTAS. PowerMacs don't
3406 	 * have a usable RTAS implementation.
3407 	 */
3408 	if (of_platform != PLATFORM_POWERMAC)
3409 		prom_instantiate_rtas();
3410 
3411 #ifdef CONFIG_PPC64
3412 	/* instantiate sml */
3413 	prom_instantiate_sml();
3414 #endif
3415 
3416 	/*
3417 	 * On non-powermacs, put all CPUs in spin-loops.
3418 	 *
3419 	 * PowerMacs use a different mechanism to spin CPUs
3420 	 *
3421 	 * (This must be done after instantiating RTAS)
3422 	 */
3423 	if (of_platform != PLATFORM_POWERMAC)
3424 		prom_hold_cpus();
3425 
3426 	/*
3427 	 * Fill in some infos for use by the kernel later on
3428 	 */
3429 	if (prom_memory_limit) {
3430 		__be64 val = cpu_to_be64(prom_memory_limit);
3431 		prom_setprop(prom.chosen, "/chosen", "linux,memory-limit",
3432 			     &val, sizeof(val));
3433 	}
3434 #ifdef CONFIG_PPC64
3435 	if (prom_iommu_off)
3436 		prom_setprop(prom.chosen, "/chosen", "linux,iommu-off",
3437 			     NULL, 0);
3438 
3439 	if (prom_iommu_force_on)
3440 		prom_setprop(prom.chosen, "/chosen", "linux,iommu-force-on",
3441 			     NULL, 0);
3442 
3443 	if (prom_tce_alloc_start) {
3444 		prom_setprop(prom.chosen, "/chosen", "linux,tce-alloc-start",
3445 			     &prom_tce_alloc_start,
3446 			     sizeof(prom_tce_alloc_start));
3447 		prom_setprop(prom.chosen, "/chosen", "linux,tce-alloc-end",
3448 			     &prom_tce_alloc_end,
3449 			     sizeof(prom_tce_alloc_end));
3450 	}
3451 #endif
3452 
3453 	/*
3454 	 * Fixup any known bugs in the device-tree
3455 	 */
3456 	fixup_device_tree();
3457 
3458 	/*
3459 	 * Now finally create the flattened device-tree
3460 	 */
3461 	prom_printf("copying OF device tree...\n");
3462 	flatten_device_tree();
3463 
3464 	/*
3465 	 * in case stdin is USB and still active on IBM machines...
3466 	 * Unfortunately quiesce crashes on some powermacs if we have
3467 	 * closed stdin already (in particular the powerbook 101).
3468 	 */
3469 	if (of_platform != PLATFORM_POWERMAC)
3470 		prom_close_stdin();
3471 
3472 	/*
3473 	 * Call OF "quiesce" method to shut down pending DMA's from
3474 	 * devices etc...
3475 	 */
3476 	prom_printf("Quiescing Open Firmware ...\n");
3477 	call_prom("quiesce", 0, 0);
3478 
3479 	/*
3480 	 * And finally, call the kernel passing it the flattened device
3481 	 * tree and NULL as r5, thus triggering the new entry point which
3482 	 * is common to us and kexec
3483 	 */
3484 	hdr = dt_header_start;
3485 
3486 	prom_printf("Booting Linux via __start() @ 0x%lx ...\n", kbase);
3487 	prom_debug("->dt_header_start=0x%lx\n", hdr);
3488 
3489 #ifdef CONFIG_PPC32
3490 	reloc_got2(-offset);
3491 #endif
3492 
3493 	/* Move to secure memory if we're supposed to be secure guests. */
3494 	setup_secure_guest(kbase, hdr);
3495 
3496 	__start(hdr, kbase, 0, 0, 0, 0, 0);
3497 
3498 	return 0;
3499 }
3500